Hermoso: AI Ad Generator MCP Server

ONE-CLICK EFFECTS, the same ones as the /tools pages: a photo goes in, a styled picture comes out, and most can then become a 5-second clip. Plush toy, 3D figure, mascot, comic book, pixel game, billboard ad, giant product, bullet time, packshot, outfit swap, headshot, relight and many more. Name no `effect` for the free list. `effect` alone: what it takes. `effect` + `images`: a free quote; add confirm:true to make the picture. Then `effect` + `still` (that picture's URL): a free quote for the clip; confirm:true makes it. `describe` adds anything of your own. For a look not on the list use generate_image. A real person's photo follows the same likeness rules as every render.

AI HOST EPISODE: format 'host_episode' = one presenter talking to camera, the camera changing every piece (frontal, three-quarter, close), the same host and set throughout, native voice, 16:9. Host = `creator` (saved or preset) or `hostImage`; words = `topic` (written for you) or `script` (verbatim). It returns a 480p DRAFT; HD (720p) is a SEPARATE call with `fromDraft` (quote it with dryRun, run it only when the user asks). Otherwise: turn a TOPIC into a finished narrated explainer video, in one of TWO LANES (`lane`). 'blocks' (the default) = 10-second VIDEO blocks, one narrated line per block, hard cuts, a music bed under the voice: an EXPLAINER renders on Gemini Omni at 720p (9:16 by default, or 16:9), a FACELESS CHANNEL video (`format:'faceless_channel'`, or channel history / kids / fairytale) on MiniMax H3 at 2K (16:9 by default) with five cuts per block. 'stills' = a picture film: writes a sectioned script, paints a BURST of pictures per section (about one every 1.5s, mostly one-detail edits of the frame before), narrates each section with TTS, holds each picture PERFECTLY STILL for its own slice of the narration (motion is the CUT RATE), then composites the end card (and any text you asked for). BURNED ON-SCREEN TEXT IS OFF BY DEFAULT; the film ships clean unless the user asks; `captions:true` adds held key points and `subtitles:true` adds subtitles timed to the narration. `style` picks the visual family: the default 'cinematic' is photoreal editorial; every other id is a STYLED, strictly non-photoreal look (illustrated / collage / clay / pixel …) that renders ONE style-key image and locks every scene to it. Cost at the default frame density: a ~130-credit hold for a 60s explainer on the default style, ~100 styled; `frameDensity:'lean'` roughly halves it and `'minimal'` (one picture per section) is ~30. All settle to the exact spend. Takes SEVERAL minutes — one image render per frame; NOT for a short product ad: use render_ad, generate_video or make_template_ad.

A reaction picture (image, or video from videoStart) with its sound, cut to when the sound lands (or seconds): a 1080x1920 clip for post_edit join.

Make a marketplace LISTING SET for ONE product in one call (Amazon, Shopify, Etsy, any store page): the MAIN image on pure white (the product alone filling about 85% of the frame, no text, no props, no added logo), then the secondary images a listing carries — a lifestyle shot, a detail close-up with NO infographic, a scale shot that shows its true size, and the product in use — every one rendered from the product's real photo and held to one visual system (the same light, scale and palette). More shots on request: multi_angle, whats_in_box, back_of_pack, or any shot in your own words. The MAIN renders FIRST and every other image is drawn from it (each waits for it in the queue), so the set matches; pass an approved main as mainImage to redo the rest from it. aplus:true adds Amazon's 7 A+ modules (a 21:9 brand hero and brand promise, 3:2 modules 2-6), each also delivered at its Amazon module's exact pixels, with its text fields suggested only from the product's own data. Priced up front: dryRun:true returns the plan and the exact credits; the charge is that number. Each image is its own queued image job (poll with get_job); they land in the Library. For a batch of ADS use make_static_ads; for one product photo in a scene use plan_ad + generate_image.

MOTION GRAPHICS and ANIMATED VIDEO in one call: kinetic typography (animated text), a logo animation or intro, a product launch film, an animated social cut. MOVING PICTURES are the default for launch, product and social: in each scene with a picture, one thing physically happens (a seed splits open, an eye opens, the view travels through a forest), filmed by an image-to-video model, with the type, transitions and the music's beat put on top; a few minutes, priced per shot, and with no pictures given each scene paints its own. moving:false, and always for type and logo, makes it in code with no video model: type, pictures and drawn graphics filmed by a moving virtual camera, a 1080p MP4 in about a minute, billed for its compute only. Pick it for 'animated video', 'motion graphics', 'kinetic text', 'text animation', 'logo animation', 'logo intro', 'animated explainer' asks. recipe: 'type' (every line its own animated card, no picture needed), 'logo' (the logo assembles out of dots and settles, the name and tagline land under it), 'launch' (30s 16:9), 'product' (15s), 'social' (15s 9:16). Or `spec`, the whole film as JSON (help:true returns the language). STEPS: without confirm it is a FREE quote that gives both prices (length, the music's tempo; the spec in structuredContent): say both; preview:true is a 1-credit contact sheet of the still version; confirm:true renders. You may write `moves` yourself, or the film's own planner writes them (that planner call is priced in the quote). Only the user's or the brand's words appear: never invent a claim, number or review. projectUrl = a past film's url returns its saved spec to edit and render again.

CONTENT FACTORY: turn ONE topic into a SERIES of faceless short videos (narration over pictures, no presenter), optionally in the STYLE of a trending reference video. Give `topic` ("make a video about…"), `episodes` (1-10, default 1) or your own `episodeTitles`, `durationSeconds` (20-120 per episode, default 30) and optionally `reference` (a TikTok / Reel / Facebook / X link or a video file URL whose look, pacing and opening hook every episode copies, with plain subtitles only if the reference burns its own; never its words or characters). Find a trending reference with search_tiktok / search_instagram / search_youtube when the user wants one. THREE STEPS, ALWAYS IN THIS ORDER: (1) call with no `confirm` for a FREE quote of the whole series (per episode and total); (2) optionally `planOnly:true` to see the episode titles and the style card first (a small planning charge); (3) `confirm:true` (pass the `plan` from step 2 to reuse it) queues every episode and returns their job ids; poll get_job. To make MORE of a series later, pass its returned plan as `moreOf`: same style, no episode repeated, the reference not watched again. Each finished episode can be scheduled with this tool again: pass `jobIds` plus `schedule` {channels, at, everyHours, timezone} and it books one post per finished episode, spaced `everyHours` apart (default 24). Captions follow the reference (on if it burns words on screen) unless `captions` says otherwise.

Make a BATCH of finished static image ads for one or several products in one call: each ad a genuinely different selling angle (offer, material or ingredient story, how it is used, a lifestyle moment, an objection, proof the brand really has) AND a different layout (editorial still life, poster headline, packshot on colour, in hand, flat lay, split compare, native phone photo and more; never one look repeated across a batch), with its headline, supporting line and CTA printed on the image and the product's real photo and the brand's real logo composited in. `products` names products from the brand's product catalog by id or name (list_products; add one from its page link with add_product; ['all'] takes every product with a photo; omit for the hero product). A product's price is printed only when it was read from its live page within the last hour: older rows are re-read first, and a page that cannot be read means no price on that ad. `count` is ads PER PRODUCT (1-20). `aspectRatios` any of 1:1, 4:5, 9:16 (each ad is composed natively once per ratio, so the ratios are different pictures with the same words; for one picture in several sizes render one ratio and use resize_ad). `brief` steers (a sale, a season, an audience) and is never turned into a price, discount or claim the brand does not state; `angles` names the angles to spread across; `layouts` (free words) replaces the layout set. ALWAYS PRICED FIRST: dryRun:true returns the plan and the quote; then pass that `plan` (and its `batchId`) back to render exactly it without planning again. Renders are QUEUED jobs, at most 12 per call: the rest comes back as `remaining`, to send again as `plan` once some finish. Read each job with get_job; never describe an ad before its URL arrives. For variants of ONE finished ad use headline_variants / resize_ad; to copy a competitor's static use clone_static.

Use this when the user asks for a quick, low-cost ad or post built from their words and pictures: a slideshow, carousel, iMessage or ChatGPT chat, Notes, kinetic type, myth vs fact, an app tour, or their own HTML design. Rendered from HTML, no AI model, ~30s, billed for its compute only, ready to post. Presets are SHORTCUTS; 'custom' is YOUR OWN design as config.html (+ css), so no layout, type, colour or motion is unsupported. custom: { html, css?, size? ('9:16' default | '4:5' | '1:1' | '16:9' | any 'W:H' | {w,h} px), durationSeconds? (1-60 = VIDEO; CSS/SVG animation and <video> are frame-stepped, scripts stripped), slides?:[{html, css?}] (2-35 = carousel) }; {{logo}} {{brandName}} {{domain}} {{accent}} fill from the brand; images and fonts load by public https URL; notes[] lists what failed to load. YOU write preset copy: short, casual, believable, finished phrases within budget. Preset ids and their fields are listed on `config`. config.music on a VIDEO: omit for a free library bed matched to its mood when the library is stocked (hermoso_capabilities hasMusic; else only its own sound effects), 'off' for silence, or words (a mood or description) to compose one for a flat fee.

Render a click-driving YOUTUBE / Shorts / Instagram THUMBNAIL or video cover through the full production pipeline (concept, casting, scene, render, tweaks, text), not a bare image prompt. Use it for any "thumbnail", "video cover" or MrBeast-style packaging ask INSTEAD of generate_image. About 9 credits per variant; the headline overlay is free. CONCEPT — open an INFORMATION GAP (the image raises a question the title answers) while staying truthful to the video. Brainstorm ≥5 concepts across the 16 frameworks (ids on `framework`; combining two is fine) before you pick; hermoso_capabilities has each one's 'realize it with' note and the emotion, overlay, font and rim-colour catalogs. THREE GATES, all BEFORE you render: 1. WHO IS IN FRAME — never assume or silently substitute a stranger. A framework with a person and no face photo is refused (nothing charged): ask the user once — themselves (a face photo, identity-locked), a generated person (`castGenericPerson:true`), or a people-free framework. 2. TEXT — default is a CLEAN render with the headline TYPESET over it (free, legible, correctly spelled): pass `headline`. `bakeText:true` only on an explicit ask for words painted INTO the image. Never infer text intent from the topic. 3. HOW MANY — ask once: one, or a SET (offer 4: one concept at different emotions / camera takes). Default 1; `variants` caps at 16. `emotion` is the biggest CTR lever on a face (identity lock is automatic for every face photo). To fix a finished one, re-call with `tweak` + `sourceImage` for a surgical edit (emotion / background / background_color / rim_light) — tweaks chain. ALWAYS check the returned postRenderCheck against the image before presenting it. PROMPT LANGUAGE — write every DESCRIPTIVE field (`sceneBrief`, `keyElements`, `location`, `composition`, `background`, `topic`, each person's `describe`, every `reference`) in ENGLISH, translating the user's words: the models render English better. `headline`, `headlineLines` and `bakedUiText` stay verbatim in the user's language.

ANIMATE A PHOTO so it talks: the presenter in `image` says `script` word for word, a separate voice lip-synced onto the still. Use it ONLY when the user asks for an animated photo / talking photo / lip-sync look by name; for any other talk-to-camera or spokesperson ask use generate_video with `speak` (the person filmed saying it, which reads as real footage). About 1-3 min, holds the pose steady, 480p/720p. The per-second credit price is in hermoso_capabilities (avatarEngines); dryRun:true returns this exact job's hold without rendering. Blocks until done where the host allows, else returns a job id to poll with get_job. Requires canAvatar. Spends credits; a refusal before rendering costs nothing.

Use this when the user asks for an image ad, ad creative, product photo or social post graphic for their business or product, or an ANIMATED GIF (format gif) or SVG file (format svg). It renders a finished image on a real image model, on their saved brand (product photos, logo), and returns a URL ready to post. For a video, use render_ad. The saved brand’s product photo is used automatically (see get_brand / list_product_photos), so never ask the user to upload it again. refImages (paths or URLs) composite a real product into the scene. MULTI-BRAND: hydration uses the SAVED brand; for a different brand (a fresh draft_brand) pass its own productImages/logo as refImages with useBrand:false, or the wrong product lands in the shot. Product photos also pick the ENGINE (the compositing model), so a named `model` is honoured only when no references ride; raw:true or useBrand:false keeps it. model = a catalog id from hermoso_capabilities (omit for the default). A PRODUCT IN A PERSON’S HANDS, or a garment on them: name an id marked `needsRefs` with a `refsMax`; person photo first, then up to three product/garment photos. It EDITS THE PHOTOGRAPH: the person is RE-POSED (stance and hands change; face, clothing, setting and light are kept), so it is the wrong tool when the rest must stay untouched. GIF / SVG run on gpt-image-2.6 / 2.6-flare from the prompt alone (no refImages); frames + frameMs shape a GIF. RAW MODEL ACCESS: raw:true dispatches your prompt to the model BYTE-IDENTICAL — no rewriting, no appended guidance, no negative prompt, no brand references attached on your behalf. Credits, the durable delivery of the finished asset and the per-model validation are unchanged. Seconds. Spends credits.

RAW music: describe it (genre, mood, instruments, tempo), get an instrumental MP3. No length: a full track at a flat fee (explainerMusicCredits in hermoso_capabilities). With seconds: by the second up to about 38 s. A named model always runs.

MAKE a sound effect or ambience that does not exist yet, from a description, as an MP3: a whoosh, a door slam, a cash-register ding, rain on a window, a busy cafe. (find_sound FINDS an existing sound, a meme or trending TikTok sound; this MAKES one.) seconds = length, 0.5-30 (default 5); loop:true for a loopable bed. engine: omit for Seed Audio 1.0 (length and loop are asked for in words, so the length is approximate), or 'eleven-sfx' (ElevenLabs Sound Effects v3: exact length, a real loop flag, description ≤450 characters). Lay it under a video with post_edit (music.track takes its link). dryRun:true returns the exact credits and makes nothing; the price quoted is the charge. Paid, priced by length (hermoso_capabilities -> sound effects).

Text generation against the writing-model catalog (Claude, Gemini, GPT, Llama, DeepSeek…) — ad copy, hooks, scripts, rewrites, brainstorms. Prompt-only, no ad assembly (for a finished on-brand creative use plan_ad -> render_ad). BY DEFAULT the model answers as a marketing copywriter (a short house system prompt is applied, which is what you want for ad copy); pass raw:true for a plain, unstyled answer from the model itself with NO system prompt at all. model = a writing-model id from hermoso_capabilities (omit for the default Claude orchestrator). Paid (billed at the writing model's token cost, by length).

Use this when the user asks for a video clip, b-roll or a shot from a description, or names a video model. For a finished video ad, Reel, TikTok ad or UGC video for their business, use render_ad. A GIF is generate_image format gif. It renders one continuous clip on a real video model and returns its mp4 URL. A clip runs up to the model’s longest listed duration (the longest takes a multi-beat 30s spot in ONE take with native audio), so never assume an 8–10s cap or stitch what fits one clip; durationSeconds must be one of the model’s durations in hermoso_capabilities. NAME A MODEL in `model` to get it: an unnamed render uses a narrower auto-pool. 1–3 min. refImage anchors the first frame; ttsScript adds a voiceover. AUDIO IS NOT FREE BY DEFAULT: a clip with no audio of its own gets a music bed composed and CHARGED on top (see musicMood and audio), which can cost as much as a cheap draft clip. refVideo (a clip URL) EDITS that clip instead, keeping its canvas and length. RAW MODEL ACCESS: by default small guards are added (label safety with no reference image, a negative prompt where the model takes one, reference lines) and hex colours become names; raw:true dispatches your prompt to the model BYTE-IDENTICAL — no rewriting, no appended guidance, no negative prompt, no brand references attached on your behalf. Credits, the durable delivery of the finished asset and the per-model validation are unchanged. Spends credits (blocked on the Starter plan).

RAW speech as an MP3: one voice reading a script (text), or a DIALOGUE of two or more people (lines), each in their own voice. Not for adding audio to a video (render_ad and generate_video voice their own spots; change_voice re-voices a finished clip). ONE VOICE: text + engine (default 'seed-audio'; also 'eleven-v4', 'minimax-speech', 'kokoro'…) + voice, a preset from that engine (hermoso_capabilities -> voice engines); a name the engine does not have is REFUSED for free with its real list, and a few engines make their own voice (the reply says which voice spoke). DIALOGUE: lines [{speaker, voice, text}] in speaking order. Each speaker keeps ONE voice, no two speakers share one, a speaker with no voice gets one nobody has; up to 10 voices and 10,000 characters, refused for free past that. Default engine: the dialogue model (Eleven v3 Dialogue, voices Aria, George, Sarah, Roger, Laura, Charlie, Callum, River, Liam, Charlotte, Alice, Matilda, Will, Jessica, Eric, Chris, Brian, Daniel, Lily, Bill, or an ElevenLabs voice id; '[laughs]'-style cues work); or name any voice engine (e.g. 'eleven-v4') to read each line on it and join them in order. dryRun:true returns the exact credits and makes nothing; the price quoted is the charge. Paid, priced by characters (hermoso_capabilities lists credits per 1,000).

Add a portrait to this workspace’s reusable CAST so the SAME person can star in future ads — the headless twin of the app’s + > Pick a creator > save. Pass the portrait’s public url (a generate_image render of an AI person, or a photo of a real person you have permission to use, or of yourself; never a photo just because it is public) plus a name to call them by; from then on list_creators returns them and their url can be re-passed to generate_avatar / generate_video / recast_motion. Saving is FREE and renders nothing. LIKENESS: leave `source` "generated" for an AI-made person (free on every plan) and use "upload"/"social" for a REAL person. Using a real person’s face, likeness or voice confirms you are them or have their consent, take full, unlimited responsibility for its use and will cover any claim against Hermoso (hermoso.ai/terms). Paid plan only; recorded.

Save a reusable PLAYBOOK — the strategy takeaways worth re-running: the hooks that work, the angles, the formats, and the concrete plays. Use it to keep what a competitor_teardown or mine_angles just found, or to bank a creative you want to repeat. Lands in the same Playbooks library the web app lists, runs and manages. Distinct from save_skill (a directive applied to every ad) and from the swipefile (raw saved creative). Free.

Save a reusable custom SKILL — a named creative directive/playbook applied to future ads (a hook formula, a UGC recipe, a compliance rule, a named specialist persona like “our founder-story style” or “short-form ad strategist”). Distill an imperative, self-contained directive. Merges into the workspace Skills library (list_skills shows built-ins + your custom skills).

Save a Hermoso render — or ANY file — into the user’s connected Google Drive. Pass a Hermoso render URL as url (or urls[] for several); for a local/external file, call upload_file first and pass the url it returns. Optional folder (created if new) + name. Returns the Drive file(s) with a webViewLink. Needs Google Drive connected (Settings > Connectors > Google Drive — one connection covers Drive, Sheets and Docs). NOTE: Hermoso uses the drive.file scope, so it reaches ONLY the files it created plus any the user explicitly handed over with the Google file picker in the app — never their whole Drive.

Save to the user's Notion as a NEW page: a research report, competitor teardown or campaign report as `content` (markdown: headings, lists, links, and an image or video URL on its own line becomes the media), a swipefile collection (swipefileCollection: name or id; each ad with its creative, copy, run dates and link), the content calendar (includeSchedule:true), or a mix. Put it under parentPageId, as a row in databaseId, or omit both for a private page in their workspace. Long text is split across blocks, never cut. The answer is read back from Notion. Free; never overwrites anything.

Save a Hermoso render — or ANY file — into the user’s connected Microsoft OneDrive. Pass a Hermoso render URL as url (or urls[] for several); for a local/external file, call upload_file first and pass the url it returns. Optional folder (created if new) + name. Returns the OneDrive file(s) with a webViewLink. Needs OneDrive connected (Settings > Connectors > OneDrive).

Save one or more ads/creatives to a named SWIPEFILE collection, creating the collection if it does not exist — the headless twin of the heart on every ad card in the web app. Use it whenever research turns up something worth keeping: a competitor ad from search_meta_ads / pull_competitor_ads, an organic post, or one of your own renders. Saved ads persist to the workspace board the web Swipefile tab shows, and feed the taste signal every future ad is planned against. De-dupes: re-saving the same ad (same key/link/media) MOVES it into the named collection instead of duplicating it. Free.

Create a folder in the user’s OneDrive (optionally nested under parentId) to organize saved files. Returns the folder id + webViewLink. Use that id as update_onedrive_file’s moveToFolderId or as parentId for a nested folder. NOTE: save_to_onedrive’s `folder` is a NAME (find-or-created), not this id.

Delete a OneDrive item — it moves to the OneDrive recycle bin (recoverable there). Pass fileId (from list_onedrive_files) + confirm:true. Confirm the exact file with the user first.

Fetch one OneDrive item’s metadata — name, type, size, modified time, a webViewLink to open it and a webContentLink to download it. Pass fileId (from list_onedrive_files). Read-only.

List files & folders in the user’s OneDrive — the root by default, a folder’s contents (folderId), or a name search (query). onlyFolders:true lists folders only. Paginate with pageToken (the cursor from a previous call). Read-only.

action list (default): every sharing link and person with access to a OneDrive file or folder: link URL, view or edit, scope, expiry, password, whether it is inherited from a parent folder. action remove: revoke one (permissionId from the list); anyone relying on it loses access at once and the link cannot be restored, so WITHOUT confirm:true nothing is removed and you get what would go. An inherited permission is removed on the parent folder, and the owner's own access cannot be removed. Removal is verified by reading the list back. 0 credits. Needs OneDrive connected.

Update a OneDrive item: rename (name) and/or move it into a folder (moveToFolderId). Pass fileId (from list_onedrive_files). To remove an item, use delete_onedrive_file.

Add a product to the brand's catalog from its product-page link: reads the page's own data (name, every photo, price and currency, description, its bullet-point selling points), no model, 0 credits. Re-adding a link updates the row (the live page wins). Without a url, `title` adds one by hand. `audience` and `sellingPoints` you pass are kept through later re-reads.

List the product photos ALREADY saved in your workspace — the brand's product library plus any app-store screens (also surfaces photos locked in your OTHER creations, since a set product lands in the shared library). FREE — returns each photo's url + the product it shows (the shop's own title when one is on file). These are already saved: never ask the user to upload them again. Call it before set_product_image to see the existing photos you can reuse. Reads YOUR saved brand (pass brandId to target a specific profile — that switches this key's active profile like use_brand).

Render an 18-30s music-led PRODUCT SIZZLE: ONE 15s Seedance 2.5 hero clip of the product, diced into fast cuts and intercut with typeset spec/CTA cards on a brand-coloured grain background, mixed to a music bed. Faceless by design — no people, no voiceover, no spoken lines; the cards carry every word, so nothing is left to a video model's spelling. Pass a real packshot as refImage or the label will not be yours. EXPENSIVE — the hero clip is the only paid leg and it is a full 15s Seedance 2.5 render: ≈870 credits at the DEFAULT 1080p, ≈360 at 720p, ≈170 at 480p; a 4k ask renders on Seedance 2.0, the only Seedance tier with 4k, ≈2,360 (call hermoso_capabilities for the live seedance-2.5 per-duration numbers; the dicing and the cards are free, and the music bed is included in these figures). Confirm the spend with the user before calling. For a talking/UGC ad use render_ad or generate_avatar; for a cheap deterministic format use make_template_ad.

Remove one product from the brand's catalog (its ads already made stay in the Library). Free.

Lock an image as the ad's real PRODUCT photo and SAVE it as this brand's default product, so every later plan_ad / render_ad / generate_image grounds on the true packaging without being told again. Pass `imageUrl` = a product shot's URL — an image from a prior research result (an organic Instagram/TikTok post, a scraped page image), a workspace / list_product_photos url, or any public product photo. The server downloads it and runs a product+safety check: a lifestyle/scene shot with no clear product, or an off-category / unsafe image, is REJECTED and NOTHING is locked or saved (the summary says why). On PASS it persists the photo to a DURABLE url, writes it to the brand's product library as the DEFAULT, and READS THE BRAND BACK to confirm — `savedToBrand` and the summary report what the brand ACTUALLY holds now, never what was asked for, so if it did not become the default you are told instead of finding out from a paid render. Bills one vision check. Reads YOUR saved brand for the category match (pass brandId to target a specific profile — switches this key's active profile like use_brand).

Edit a catalog product's title, description, selling points, audience or page link. What you set here is kept through later page re-reads. Free.

Create a folder in the user’s Google Drive (optionally nested under parentId) to organize saved files. Returns the folder id + webViewLink. Use that ID as update_drive_file’s moveToFolderId or as parentId for a nested folder. NOTE: save_to_drive’s `folder` is a NAME, not this id — it find-or-creates a folder by that name, so pass the folder NAME there (or omit and just save, then move with update_drive_file).

Delete a Drive file. By default it goes to Trash (recoverable); pass permanent:true to delete it forever. Pass fileId (from list_drive_files) + confirm:true. Irreversible when permanent — confirm with the user first.

Fetch one Drive file’s metadata — name, type, size, modified time, a webViewLink to open it and a webContentLink to download it. Pass fileId (from list_drive_files). Read-only.

List the Google Drive files & folders Hermoso can reach: the ones it created plus any the user picked with the Google file picker in the app (drive.file, never their whole Drive). This is how you find a picked file's id. Filter by query (name contains …), folderId, or onlyFolders:true; paginate with pageToken. Read-only. A FOLDER: picking a folder hands over the folder alone, not its files, so a folder can list empty or partly; the reply then carries `pickUrl`, an app link that opens the picker inside that folder (Cmd+A / Ctrl+A selects every file and subfolder). Give the user that exact link, then list again. recursive:true walks every subfolder Hermoso can see (each row has its `path`) and reports picked files whose folder it cannot see.

Update a Drive file: rename (name), move it into a folder (moveToFolderId, optionally removeFromFolderId to move OUT of the old one), or trash / untrash it (trash:true|false). Pass fileId (from list_drive_files). To delete permanently, use delete_drive_file.

Add a NEW profile (a brand, client, creator or personal workspace) and switch to it. Each has its OWN brand details, memory, swipefile, Library, avatars, skills, playbooks and connectors; nothing leaks between them. Use draft_brand to FILL it and update_brand to edit it. Re-running with the same name returns the existing profile instead of a duplicate. Creating it is free. Drafting a NEW brand into it (draft_brand) needs the credits for its research on the balance, about 100; with fewer, draft_brand refuses before anything runs or is charged and says how many it needs and how many the account has.

PERMANENTLY delete a profile and EVERYTHING in it — brand details, memory, swipefile, Library creations, generated assets, avatars, skills, playbooks, chats — and disconnect its connected accounts. Irreversible, and it applies to everyone the workspace is shared with. Call it WITHOUT confirm first: it reports exactly what that workspace holds. Show the user that inventory verbatim, get an unambiguous yes, then call again with confirm:true — plus, if the workspace is not empty, confirmName set to its exact name and confirmConnectors set to the number of connected accounts it reported. Those two exist because confirming INTENT does not prove you picked the right WORKSPACE, and a wrong target is how a live profile was destroyed. The account's FIRST/anchor profile cannot be deleted this way (it holds the workspace's root storage) — that one is replaced from the app.

What Hermoso ALREADY KNOWS for the active profile: the same saved profile (products, logos, palette, positioning) + learned memory the web Studio uses. Call it when you need to know whether a brand is on file: if hasBrand is true you can omit brand everywhere; if false, onboard with draft_brand. Not a required first step before a render: the create tools read the saved brand by themselves. 0 credits.

Patch SPECIFIC fields of the active profile (name, domain, sells, summary, category, audience, positioning, voice, style, goal, logos, brand/product name pronunciations, profileType: explore = kept but unused) WITHOUT overwriting the rest — a read-modify-write on the saved brand. Use for “change our voice to playful”, “we sell to dentists now”. To onboard a brand from scratch, use draft_brand. If no brand is saved yet and you only INFERRED one from what the user is making, ask them to confirm it is their brand before saving it (they may be working for a client or just trying things). Only pass the fields you’re changing.

Find the Notion pages and databases the user shared with Hermoso, by title, newest edited first. Gives the page id that save_to_notion, append_to_notion_page and read_notion_page take, or the database id to save rows into. Read-only, free. Needs Notion connected (Settings > Connectors > Notion).

Read the ROWS of a Notion database the user shared with Hermoso — e.g. the team's content calendar: each row's title, id, link and every property as a plain value (status, date, channels, caption, links), plus the database's columns with their types and options. Filter with match {"Status": "Scheduled"} (typed from the database's own columns) or a raw Notion filter object; sort with sortBy (a column, or created_time / last_edited_time) and direction. Up to 100 rows per call, cursor for more; the answer says when Notion stopped at its 10,000-result limit. Use it to import a calendar and schedule its posts, or to find the row update_notion_row changes. Read-only, free. Needs Notion connected (Settings > Connectors > Notion).

Read a Notion page the user shared with Hermoso, as markdown (pageId or a Notion link). A database id answers with its data sources instead. Read-only, free.

Change properties on ONE row of a Notion database (pageId from query_notion_database) — e.g. mark a calendar row Status "Posted" and put the live post link in its URL column. properties is {"Column": value} by column name: text, number, checkbox, select (an existing option or a new one), status (must be an existing option), multi-select (array; replaces the whole value), date (ISO, or {start, end}), url, email, phone; null clears. Validated against the database's own columns before anything is sent and refused by name otherwise, never coerced; columns you do not name keep their values. Read back from Notion. trash:true moves the row to Notion's trash (restorable; returns a preview first and needs confirm:true), trash:false restores it. Free.

Create a new Google Spreadsheet in the user’s Drive and optionally fill it with rows — e.g. export a swipefile, ad list, or performance report. Pass rows as an array of row arrays (first row = headers). Returns the spreadsheet id + URL. Needs Google Drive connected (Settings > Connectors > Google Drive — one connection covers Drive, Sheets and Docs).

The tabs in a Google Spreadsheet, each with its name, numeric sheetId, row/column count and position. Call this BEFORE naming a tab in update_sheet / clear_sheet_range / manage_sheet_tabs / format_sheet, and before proposing to delete one — it is how you learn what the file actually contains instead of guessing at a name. Read-only, free.

Read cells from a Google Sheet Hermoso can reach — one it created, or one the user handed over with the Google file picker in the app (that is how an EXISTING spreadsheet becomes readable; find its id with list_drive_files). Pass the spreadsheetId (from create_sheet) OR paste a Google Sheets URL as sheetUrl. If Google answers that the file was not found, the user has not picked it yet — ask them to pick it in the app rather than retrying. Returns a 2-D array of values.

CORRECT cells in a Google Sheet — write values to an exact range, overwriting whatever is there. This is the fix append_to_sheet cannot make: appending only ever adds rows at the bottom, so without this a wrong number stays wrong forever and the only "correction" is a second row contradicting the first. Pass `range` (e.g. "B2:C5", or "Q3 Report!B2" to name a tab — list_sheet_tabs gives the names) and `values` as an array of row arrays; an anchor cell like "B2" is fine and the block is written down and right from it. Writing into EMPTY cells goes straight through. Writing OVER cells that already hold values is REFUSED first, naming exactly how many filled cells would be overwritten — show the user that, get a yes, then call again with confirm:true. The result is READ BACK from the sheet, so what you report is what the sheet now holds rather than what Google accepted.

Append text to the end of a Google Doc Hermoso can reach — one it created (pass the documentId from create_doc) or one the user handed over with the Google file picker in the app (find its id with list_drive_files). Optional `dropdown:{title, options:[2-50], selected}` appends a Google Docs dropdown chip after the text — e.g. text "Status: " + dropdown {title:"Status", options:["Draft","In review","Approved"]} gives a brief a status the team can click to change; the chip is confirmed by reading the doc back. Set it later with update_doc dropdowns.

Add to the END of an existing Notion page (pageId or its link, from list_notion_pages): markdown content, a swipefileCollection, or includeSchedule:true. Nothing already on the page changes. Free.

Append rows to a Google Sheet Hermoso can reach — one it created (pass the spreadsheetId from create_sheet) or one the user handed over with the Google file picker in the app (find its id with list_drive_files). rows = array of row arrays.

Create a new Google Doc in the user’s Drive with a title + optional body text — e.g. export ad copy, a creative brief, or a report. Returns the document id + URL. Needs Google Drive connected (Settings > Connectors > Google Drive — one connection covers Drive, Sheets and Docs).

Read the text of a Google Doc Hermoso can reach — one it created, or one the user handed over with the Google file picker in the app (that is how an EXISTING doc becomes readable; find its id with list_drive_files). Pass documentId (from create_doc) OR paste a Google Docs URL as docUrl. Under the drive.file scope it reaches nothing else in the user’s Drive; if Google answers that the file was not found, the user has not picked it yet — ask them to pick it in the app rather than retrying. Returns the plain text of EVERY tab (each headed by its tab name when there are several), with smart chips (dropdowns, people, dates, rich links) shown as the text Docs displays, plus `dropdowns[]` — each chip with its title, current value and options, which is what update_doc dropdowns sets. Read-only, free.

EDIT a Google Doc — the correction append_to_doc cannot make. Two shapes: `replacements:[{find, replace}]` rewrites specific text wherever it appears (call read_doc first and match the text EXACTLY; matchCase:false ignores case), or `rewrite:"…"` replaces the ENTIRE body (rewrite:"" empties it), or `dropdowns:[{title, value}]` sets a dropdown chip (e.g. Status -> Approved — read_doc lists every chip with its options; pass dropdownId when two share a title; an unknown title or option is refused with the real list and nothing changes); the same entry with options:[…] (the COMPLETE new list, 2–50) and/or newTitle changes the dropdown itself for every chip built from it, and an option you drop that a chip shows needs replace:{"Old":"New"}. Find/replace runs immediately and REPORTS how many occurrences changed — zero matches is reported as a FAILURE to match, never as a quiet success, because a text edit that silently does nothing is worse than one that visibly fails. A whole-body rewrite is destructive: call it without confirm first to get the character count, then confirm:true + confirmCells. Both are index-free by design — an agent cannot reliably compute Google’s character offsets, and a wrong offset deletes the wrong sentence.

EDIT an existing image in place with a plain-language instruction and keep everything else: 'make the headline bigger', 'add our logo bottom right', 'swap the background for a kitchen', 'remove the person on the left', 'erase all the text'. Pass `image` (URL, Library item, upload_file URL or local path) and `instruction`. The same edit the web Studio's Edit runs: composition, aspect ratio, people and every untouched line of text stay as they are; the saved brand's real name and website are pinned so an added line never invents one, and the brand's real logo FILE is put on when the instruction asks for the logo (never redrawn by the model): an ask only about the logo ('the logo is wrong', 'make the logo bigger'; or logo:true, any language) runs no image model. Set removal:true when the edit STRIPS text, branding or an object, so nothing branded is put back. When the saved brand has a product photo and the ad shows that product, the photo rides with the edit and the product's label is read and checked against it: re-printed from the photo only when it came out wrong (the check alone is charged when it is right; one small extra charge finds the product first). fixLabel:false turns that off. For a precise region, pass `mask` (see generate_image). One image edit's credits; returns the new image URL. For a new image from a prompt use generate_image; to rebuild a competitor's ad for your brand use clone_static.

Compose ANY edit or transition yourself; there is no preset list. Segments on an output timeline (later ones drawn on top; overlapping ones ARE the transition), each animated by keyframes, plus your own HTML overlays and masks, compiled server-side into one render. Whip pan, zoom through, push, spin, speed ramp, slow motion, freeze frame, reverse, flash, J-cut / L-cut, circle or shape wipe, split screen, picture in picture: all composed from the fields below. Local render, the flat post_edit price (~2 credits, overlays included); a `generate` segment (a paid generated transition shot) only when the user explicitly asks for one. LOOK FIRST AND AFTER: video_frames(url, start, end, fps) shows exact frames free, so find the moment (the second the egg tips) before you cut, and check the seam of what you rendered before you call it done. SOURCES: src is 'this' (videoUrl) or any Library / render URL, direct file or PUBLIC TikTok / Reel / Facebook / X post link (a viral hook's link works as is; a YouTube link is refused, YouTube does not let us download videos, so upload the file instead). in/out are SOURCE seconds; out:'payoff' cuts just before the hook's payoff, found from its motion. AUDIO: audio.tail:'payoff' carries the hook's OWN payoff sound across the cut (the egg's real splat over our clip); tail seconds = an L-cut; lead = a J-cut; gain dB or keys; mute. A VIRAL HOOK + THEIR PRODUCT: start from the hook. A clip matched to an unrelated hook never reads as one video, so write the clip AFTER the hook for it: a linking script + shot brief (the first line answers the hook, e.g. 'still waiting for the egg to land... anyway, come check out our restaurant'; what to film so it follows on; 5-15 s; then the pitch). The user records it, or you generate it (render_ad / generate_video, cost quoted first, on their OK); then join here: the hook with out:'payoff' and audio.tail:'payoff', then their clip. Match an existing unrelated clip only if they insist. A {generate:{prompt, seconds 3-8}} segment (a generated transition-only shot, paid, postEditTimeline) is never suggested; build it only when they explicitly ask for one. LINK FIRST: an effect alone never connects two unrelated clips; the link comes from what is in the frames. (1) Match cut, the default: video_frames (with its MOTION readout) on the hook's last second and across the other clip; pick the out-point AND the in-point (in: seconds, not always 0) where a motion direction, a screen position or size, a shape, a surface, a gesture or a gaze carries across, then ride the effect on that shared motion. (2) Its host names the hook in the first line. If the two share nothing, say so and offer the follow clip made for the hook. PRO, NOT IMOVIE: ease every curve (never linear on a move); keep the picture filling the frame through a move (scale up while it moves: two frames sliding side by side with a seam is the amateur tell); hide the handoff under the fastest, blurriest frames; carry direction into the next shot; cut on motion; end every effect cleanly; 0.2-0.6 s in total; a sound whose peak lands on the handoff (sfx whoosh at handoff minus 0.45 s, or the hook's own payoff sound). Moving segments get a real shutter blur automatically (motionBlur). EVERY SEAM IS MATCHED AUTOMATICALLY, hard cuts too: each cut is measured and the incoming clip graded (exposure, white balance, black level), grained UP (never smoothed) and softened while it moves toward the outgoing one; the reply gives before/after deltas per seam. match (timeline: every cut; segment: the cut into it): 'auto' default, 'off' for a deliberate contrast, or {grade, level, grain, blur: false to skip one, strength 0-1}; a segment's own constant exposure / contrast / saturation replaces the automatic grade. EVERY STITCH CHANGES THE SHOT: where two different clips (or two non-contiguous moments of one) meet at the same framing, the incoming one gets a simple crop of the same footage, varied: zoomed in on the subject, back out to the wide, or reframed off-centre (reframe: 'off' or the step 1.1-1.5; a segment with its own framing, a card or a transition is left alone). Still yours: subject size and headroom (scale it, never a jump from a third of the frame to two thirds) and sound (a 0.25-0.5 s J/L-cut, never a sonic wall). A clip placed after a hook gets a BUDGET (total - intro = its surviving window, and what was dropped). The plainest thing that links wins: a straight cut on action beats a decorative effect; over 0.5 s is too long in anything under 20 s; never flash more than 3 times a second. RECIPES (c = the cut second, adapt freely): whip pan: A over its last 0.22 s x 0 to -0.22, scale 1 to 1.35, mblur 0 to 220, all ease in; B overlap 0.08, opacity 0 to 1 over 0.08, x 0.22 to 0, scale 1.35 to 1, mblur 220 to 0, all ease out over 0.3 s; whoosh at c-0.45. Zoom through: A over its last 0.35 s scale 1 to 3 ease in anchored on the object, blur 0 to 10; B overlap 0.12, opacity 0 to 1, scale 1.5 to 1 and blur 10 to 0 ease out over 0.4 s. Cut on action: A out ON the motion, B scale 1.08 to 1 ease out over 0.25 s, audio.lead 0.2. Speed ramp: speed [{src:t0,v:1},{src:t0+0.25,v:0.3}] then [{src:t1,v:0.3},{src:t1+0.1,v:2}] into the cut. Circle wipe: B overlap 0.5 + overlays [{mode:'mask', segment:1, start, end, html: a white div whose clip-path circle grows via @keyframes}]. Card (picture in picture: a proven ad playing in a rounded card over the host watching it, any length): the host segment full frame, then the clip ON TOP with at:0, fit:'contain' (crop to reframe it), scale ~0.6-0.7, y ~0.12, radius ~0.04-0.06; a card is not a cut, so it is never graded toward the host; duck it under the host's first line with audio.gain keys (the host's voice leads the switch) and end it on a hard cut at a sentence break. SELF-CRITIQUE: the reply carries a vision REVIEW of each seam (pro / ok / amateur, linked or not, with fixes; about 2 credits, review:false skips it) and the seam frames. When it says ok or amateur, fix what it names and re-run the same sources (twice at most) before presenting; on a re-run give a generated segment {src: its URL, between: true} so it is not paid for twice. Then look at the WHOLE result once with video_frames, not only the seams: the first frame is not black or frozen, no dead air over ~0.3 s at the head, no lone black, flash or repeated frame at a cut, and nothing static for more than ~4-5 s (recut it or add a re-hook). FOLLOW-UPS ('cut earlier', 'no splat', 'whip pan instead', 'use the second hook') re-run this with the SAME sources and the one change; the result echoes the resolved timeline (e.g. the found payoff cut) to edit from. Refusals are free and name the field.

SWAP AN OBJECT IN A VIDEO ('replace the cup with our can' + the product photo as elements) or EDIT an existing clip from a plain instruction (video-to-video): the motion, timing, framing and cut stay, the named thing changes. 'change only the mug to red', 'make it nighttime', 'restyle it as claymation'. Your words are wrapped so the model keeps everything else identical, changes only what you named and repeats that lock; lighting is kept unless the change needs new light (set lighting to force either); literal:true sends your words as written. One change per call holds best. previewFirstFrame:true edits ONE still first (one image edit; previewAt picks the second) and quotes the clip; nothing else runs until you call again, ideally with previewStill. The reply scores how well the shot held outside the change (free) and flags an edit that touched more than asked. NOT for cuts/trims/end cards (post_edit), a new video (generate_video / render_ad), translating what is SPOKEN (dub_video) or a saved creator's face (recast_motion). Translating or rewriting the on-screen words of a video laid out from a design (a template ad, a custom design) is textChanges with every line at once: one re-make of the design, about 2 credits. Best on 3-15s clips.

List the connected Facebook Page's or Instagram account's OWN existing posts — id, caption, permalink, publish date and format. Use it for "how did our last few posts do", to find a post the user describes loosely, or before backfill_posts. Pass target:'instagram' for the linked IG account (Stories are excluded — Meta's media edge does not return them); Facebook hides unpublished drafts unless you ask for them. Only ever reads a Page the brand has connected. Read-only, 0 credits.

Structured Meta (Facebook/Instagram) Ad Library pull — use when you know exactly WHAT to fetch: a keyword (query) OR one advertiser (companyName / pageId). Returns compact JSON {page_name, body, cta, link, dates, media} per ad. For open-ended research that needs judgment across platforms, use research_ads instead. Spends credits on each ad-library read it makes: one for a keyword; for a brand name, the page lookup and its ads (plus a keyword search when that page has none).

Find a PLACE to tag on an Instagram or Facebook post: a shop, cafe, restaurant, venue or landmark. Give its name with the city ("Blue Bottle Coffee Oakland"), or paste its Facebook Page link, an Instagram place link (instagram.com/explore/locations/...) or its numeric id. A pasted link or id is read and comes back as that Page's taggable pageId with its name and address, or a plain reason it cannot be tagged. Returns a handful of matching Facebook Pages that have an address: pageId, name, address, category, link. These are matches to CONFIRM with the user (check the name and address; a place tag is public), and Instagram checks the place when the post is made. Needs no connected account. Spends about 1 credit per Page it checks (a name search checks up to 5 Pages, so about 5 to 7 credits); a pasted Facebook Page link or numeric id is about 1 credit; an Instagram place link costs nothing.

Plan an ad: turns a brand and a product or brief into an ad concept, with copy variants (headline, primary text, call to action) and either an image prompt or a timed video storyboard, plus the recipe and the model ids to render with. It renders nothing; pass its output to render_ad, generate_image or generate_video. When the user names a length ("a 30 second ad", "make it 45s"), pass it as durationSeconds: the storyboard is written to that length and render_ad renders it as one clip or as joined clips. Without it the planner chooses a length that suits the format, so a length the user asked for belongs in durationSeconds rather than only in the brief. The user’s own screenshots or pictures to show inside the video go in `inserts`. The brand’s SAVED product photo (get_brand / list_product_photos) is used automatically; never ask the user to upload it again. Uses credits for the planning step.

Fan a brief into N DISTINCT ad angles (different hooks/mechanics/audiences), each with its own headline + visual brief — then render each with generate_image and rank with score_ad. LLM planning only; renders nothing itself. To plan AND render a batch of finished static ads for one or several products in one call (up to 20 per product), use make_static_ads.

Use this when the user wants a week of social posts planned, or answers yes or no to planning their week. It drafts one batch of 5 posts for the active brand: a new image, a caption and a suggested day and time for each. Without confirm it returns the price and whether the batch fits the account’s posting allowance, and spends nothing. Confirm:true starts the batch and bills credits; read the posts with get_job. Decline:true records that the user said no.

ONE finished static ad in, VARIANTS × FORMATS out, in one job, as a named file list (<name>_<variant>_<format>_<WxH>.png) with a manifest. The ad's text is read ONCE as the source of truth; each variant's change is made ONCE on the master; every format is derived from that master; every output is text-checked line by line and any line it lost is named (never re-rendered). Variants: {headline}, {language}, {swap:[{from,to}]} (exact words: a price, an offer, a CTA) or {edit} (any change in words), each with an optional {name}; the original is always included. Formats: default 1:1, 4:5, 9:16, 16:9, or 'everywhere', any canvas or a platform word (snapped to the nearest canvas an image model makes). At most 24 renders. dryRun:true = the exact price, which is what it charges. One resize alone is resize_ad.

Turn a VIDEO into WRITTEN pieces: a summary, YouTube chapters (first at 0:00, at least 3, each 10s+), chapter summaries, an X thread (every post under 280 characters), a blog post, timestamped verbatim quotes, or any piece in your own words (pass the words as a format). The video: a YouTube link (read from its captions), a video file URL, a Library id, an upload_file URL or a social post link (read by watching it); or the text itself, when you already have it from analyze_video (which is the tool that just reads a video back). Several formats share ONE read. Priced by the video's length before anything is spent: without confirm it returns the exact credits and spends nothing; confirm:true writes and charges exactly that. Only the paid minutes are read. Every piece is written only from what the video says. Built-in skills youtube_script and retention_skeleton (get_skill) write long-form scripts.

ONE video in, EVERY placement out, in one job, as a named file list (<name>_<placement>_<WxH>[_<lang>].mp4) with a manifest. Placements: 'everywhere' (default: 9:16, 4:5, 1:1, 16:9), any W:H, a pixel size (1200x628) or a platform word (tiktok, reels, stories, feed, square, youtube, linkedin, pinterest). Each is reframed by a BLUR-PAD by default (the whole frame over a blurred copy of itself; free, no AI), reframe 'crop' (fill, edges cut; free) or 'ai' (the new canvas generated around the frame; paid, for a canvas the engine makes, else padded and said). A placement may be {placement, reframe} to mix. Optional captions (word-timed, ~4 words a cue, placed for each canvas) and languages (a lip-synced dub per language; every placement of each). A pack of pads and crops only is billed 0. dryRun:true = the exact price (the sum of every leg's own price), which is what it charges. For ONE canvas use post_edit op pad.

One-click STATIC-AD CLONE (the web app calls it Clone): rebuild a competitor/reference STATIC (image) ad as an on-brand version — SAME layout, composition and energy, but YOUR product, brand colours, logo and voice, with every trace of the source brand removed. Pass `imageUrl` = the static ad image to clone. Uses your saved brand (pass brandId to target a specific profile — that switches this key's active profile like use_brand). IMAGES ONLY — for a video ad use clone_video with its link, then render_ad. Bills as one image generation.

Remake a video you like FOR THIS BRAND from its link — a TikTok, Instagram Reel, Facebook video or reel, X post, or a direct video file URL. YouTube does not let us download videos, so a YouTube link is refused up front and nothing is charged; the user downloads it themselves and uploads the file (upload_file). Hermoso WATCHES it first (frames across the whole clip plus a transcript of the voiceover, on-screen text and cut map), then plans a storyboard that keeps its hook device, structure, jump cuts and pacing while swapping in THIS brand's product, cast, setting and words — never the original's words, face or brand. The new ad MATCHES THE ORIGINAL'S LENGTH (capped at 60s) unless durationSeconds is given. Renders nothing: pass the returned creative to render_ad to make the video. Costs the plan plus about 2 credits to read the link. The reply says exactly what was watched, and when a platform will not hand over the footage (YouTube sometimes refuses servers) it says the plan rests on the captions and thumbnail only. For a local file, upload_file it first and pass the URL.

Tear a competitor's ad strategy down into an actionable playbook: their opening-hook MIX, longest-running campaign THEMES, the WHITE SPACE nobody in their set runs, 2-3 render-ready COUNTER-PLAYS, and the territories they own that you should avoid. Pass `competitor` {name, domain?}. CONTRACT: supply `ads` (raw ad objects from a prior pull_competitor_ads / search_meta_ads call) to tear exactly those down, OR omit `ads` and this pulls the competitor's real Meta ads first (spends the same ad-library reads search_meta_ads does, longest-running = proven winners). Auto-tailors the white space + counter-plays to YOUR saved brand. Spends credits (free when you pass ads).

Set, change or stop a brand's MARKET RADAR (the standing competitor watch; Ad Spy > Watching in the web app; each brand has its own), run weekly or daily: newAds (what watched brands launched), winners (their ad passing 30 days running), trends + topPosts (formats and most-viewed posts for niche topics), ideas (3 to make). Only what you pass changes. competitors / topics REPLACE their lists (max 8 brands, 5 topics); both empty stops it and clears findings. auto:true picks 3-5 competitors, 2-3 similar brands and 2-3 topics from the saved brand or `seed` (one model call, billed at its token cost). dryRun:true quotes one check, saves nothing. Saving is free; each check holds its quote first and is skipped when the balance is short. runNow is refused while a check runs or just after one finished. Read it with list_watch_findings.

Show every identity a connected account can act as, and which ones this PROFILE is currently allowed to use — Facebook Pages + Instagram + Meta ad accounts, Google Ads customers, LinkedIn company Pages (and the personal profile), Pinterest ad accounts, Microsoft Advertising accounts. One person often administers several; only the ticked ones can be posted to or spent from. Call this before set_connector_accounts, and let the USER pick — never guess. Providers: tiktok, x, youtube, threads, bluesky, telegram, pinterest, instagram, meta, google_ads, linkedin, pinterest_ads, linkedin_ads, reddit_ads, apple_ads, microsoft_ads, google_business, google_analytics, snapchat_ads, x_ads, tiktok_ads, google_tag_manager, google_search_console, bing_webmaster. Read-only, free.

Set WHICH of a connector's accounts this profile is allowed to post to and spend from — Facebook Pages / Instagram / Meta ad accounts, Google Ads customers, LinkedIn company Pages (and the personal profile), Pinterest or Microsoft Advertising ad accounts. Pass ids from list_connector_accounts. This REPLACES the current selection: anything you leave out is un-shared, and an EMPTY list shares nothing (publishing then refuses — it fails closed by design, and the server re-verifies every id against the live connection, so an id the account cannot actually reach is rejected rather than saved). Ask the user which accounts they mean; posting as the wrong Page is a public mistake. Providers: tiktok, x, youtube, threads, bluesky, telegram, pinterest, instagram, meta, google_ads, linkedin, pinterest_ads, linkedin_ads, reddit_ads, apple_ads, microsoft_ads, google_business, google_analytics, snapchat_ads, x_ads, tiktok_ads, google_tag_manager, google_search_console, bing_webmaster. Free.

Scan organic TikTok, Instagram Reels and YouTube for a niche across a few query variants, fold the posts into creators, and rank them on median views, engagement rate and how often they show up for that niche; the top rows get follower counts AND public contact info (an Instagram business email / phone / category, the bio link, an email in a TikTok bio) so outreach can start from the result. Real people, not AI actors — for influencer sourcing, UGC casting and partnership prospecting ("who should we send product to?"). About one credit per search call (platforms × queries, default 3 × 3) plus one per enriched profile; repeats inside 20 minutes are free. Then shortlist (save_to_swipefile), check a profile, draft outreach (generate_text), or approve them for Partnership Ads. Marketplace:true also searches Instagram’s creator marketplace (Meta’s own creator directory: followers, badges, marketplace email) for the same niche and returns those rows beside the ranked list.

List this workspace’s SAVED CREATORS — the reusable on-camera cast (AI creators made here, a person pulled from a social profile, a consented photo upload). Read-only, FREE. Each entry gives the name, the PORTRAIT URL, where the portrait came from and whether a real person’s likeness consent is on file, how many extra pose plates exist, and any chosen or cloned voice. TO PUT ONE IN A FINISHED AD, pass their id or name as render_ad’s `creator` — that casts them for the whole spot (and skips the character-portrait render, so it costs less than not casting anyone). THE PORTRAIT URL IS THE REUSE HANDLE for the raw lanes — pass it as generate_avatar’s `image` (a talking clip of them), generate_video’s `refImage` (they star in the scene), recast_motion’s `image` (they perform a reference clip’s motion), or generate_image’s `refImages`. CALL THIS BEFORE OFFERING TO GENERATE A NEW PERSON: re-casting somebody the workspace already has keeps the SAME face across every ad, while a fresh person costs credits and breaks that continuity. An empty answer means the workspace genuinely has no cast yet — say so and offer generate_avatar / save_creator, never invent a roster.

Export the posting schedule to a Notion database, one row per post: date, channels, caption, media link and status. Creates a new database (under parentPageId, or private in the workspace) or writes into an existing one (databaseId), matching its columns by name. historyDays adds posts already published. Up to 100 rows per run; the answer says if more were left. Free.

Turn a SWIPEFILE COLLECTION into a real Google Slides deck — one slide per saved ad, carrying the creative, the brand, the ad copy, the run dates with the run length, and the platform. This is the thing a marketer actually presents to a client or a team. Returns the presentation id + URL. Creates a NEW deck every time: under the drive.file scope Hermoso can only touch files it created, so it cannot add slides to a deck the user already has. A creative whose ad-library link has expired cannot be embedded — Meta signs those URLs with a short expiry — so that slide says so in words and keeps its copy and run dates, and the reply reports how many. Needs Google Drive connected (Settings > Connectors > Google Drive — one connection covers Drive, Sheets, Docs and Slides).

FINISH a 480p Seedance 2.5 DRAFT in 1080p: the model regenerates THAT take (same scene, shots, motion, sound) at full resolution. Every 480p Seedance 2.5 render (generate_video / render_ad at resolution 480p) is a draft, finishable for 7 days; its result carries `draft` {id, expiresAt, finishCredits}. Pass that id, or the draft video URL. It is the full-resolution route for a draft (upscale_video is for a video that is not a draft); priced like a 1080p render of its length. Paid: call with dryRun first, state the credits, run only when the user agrees. Captions, end card or music added after the draft are not re-applied. Not a draft or expired: a plain refusal; use upscale_video or a fresh 1080p render.

Post-process an EXISTING rendered video (its served mp4 URL) with the proven direct-response 'reviewer' finish and/or a film-grain pass — no AI model, ~30s, billed for its compute only. pills=true composites a header pill (e.g. '10/10 would buy again'), a brand-accent sub-pill, and 3-4 green-check proof pills cascading in on the beat (YOU author the copy: header ≤40 chars, sub ≤34, each point ≤44 — concrete real benefits, never fabricated stats). grain=true applies a subtle camera-grain finish that makes photoreal AI renders look phone-shot ('less AI') — works alone or with pills. Returns a NEW video; the original is untouched.

Probe what this Hermoso account can do RIGHT NOW: available image/video model ids + their credit costs (estimates: a render is charged its actual cost), aspect ratios, video durations, the recipe ids, and the canEdit/canAvatar flags. Call it when you need a specific model id, a credit cost, or a capability you are not sure of. It is NOT a prerequisite for rendering: generate_image, generate_video and render_ad all run with `model` omitted and route to the server’s own default. Read-only, free.

Return the credit balance, recent priced calls with their credits, and costModel (the one-sentence rule of what costs credits). THE RULE: only AI model runs and Ad Spy research spend credits; publishing, scheduling, ads management, analytics, comments, DMs and connectors cost no credits on any plan (X is the single per-call exception; xCost gives its exact credits per post, with and without a link). Check before kicking off paid generation; answer "does posting cost credits?" with NO. HOW MANY posts is set by the plan, not credits: the Free plan includes 10 posts a month on up to 3 channels, and a post counts once however many channels it goes to (`posting` says how many are used); every paid plan posts without a limit and runs autopilot (Basic $9.99/mo: unlimited posting and Autopilot, no monthly credits; it includes the same 250+ free credits as the Free plan, monthly credits start on Starter, and Autopilot spends credits). Starter, Pro and Studio: media models at cost + 10% (memberRate); Basic and Free pay standard prices.

PERMANENTLY delete images/videos from this brand's Library: the entry and the stored file. Pass ids (L0… from list_library) or urls. Without confirm it lists what would be deleted; ask the user, then call again with confirm:true. Free.

Browse this workspace's Library — every image/video generated in the Studio, and uploads (source:'upload'), newest first (the same Library the web app shows). Returns served URLs you can open directly or hand to fetch_asset for a download link, plus each asset's id (L0, L1…, which plan_ad and render_ad `inserts` take), a human TITLE (the ad's headline or concept, the prompt's first words, or the upload's name), kind, model, and age. When the user names an ad by its headline ("the Afternoon fix ad"), pass `query` to find it by title. Free, read-only.

MECHANICAL post-production on an EXISTING video (URL): ordered primitives run by ffmpeg, ~2 credits flat, NO AI model, as a NEW video. Ops: a branded end card, trim, speed (0.5-2x), mute (whole or a window), audio_gain (-20..+6 dB), fade_out, music (a bed under the clip, ducked: omit track = a free library track picked by mood, no attribution; or track = an audio link: an upload_file URL, a find_sound or generate_sfx link, a video post's sound; track 'generate' composes it, paid, ONLY when the user asks; start/end, db, replace), voiceover (text spoken OVER it, its sound ducked; ~2.5 words/s; voice female|male; +~1 credit), watermark (brand logo), grain (anti-AI), text (timed words in a native look: style 'tiktok-classic' default / 'clean-minimal' / 'note-style' or a textStyle; start/end, or at:{word} and until:{word} to time them to a spoken phrase), join (this video FOLLOWED BY clips[]: Library URLs, direct files or public post links, as one 1080x1920 video). Presets are shortcuts: x/y, grain amount, join transition and bridge sound are free parameters. 'A viral hook, then our clip' = videoUrl: the hook's post link + [{op:'join', clips:[{url: ours}], bridge}], and it ALWAYS gets a bridge unless the user asks for a bare cut: {kind:'impact'} cuts the hook just before its payoff (cutAt overrides) and lands our clip on a punch-in and flash, with the payoff sound FROM THE HOOK ITSELF: its own audio carries across the cut, else one generated from its frames (up to 8 credits), else a neutral impact. {kind:'text', text, then?} only when the user asks for words over the cut. ANY OTHER EDIT, op or bridge (whip, zoom, freeze, wipe, split screen, generated shot) is edit_timeline: free keyframes, any size. NEVER generate_video/render_ad for these. Feedback maps to one named move, never a re-render: quieter/louder (music db, audio_gain), faster/slower (speed; slower = zooms and pans at 0.7x), tighten (trim), retime (to a word anchor), recut (order, in-out points), regrade (timeline exposure/contrast/saturation keys).

Use this when the user asks which of their social posts did best. Aggregate this brand's published posts to answer WHICH HOOKS AND SUBJECTS WORK. Groups by hook (default), subject, format (recipe), channel, media format or posting hour, reports the engagement RATE within each channel, and ranks the best and worst POSTS in each channel. Describe a post by the creative it carried (what it shows, its format, its link), not by its caption — the caption is the least important part of a post. THREE THINGS IT DELIBERATELY WILL NOT DO, and you should repeat them rather than paper over them: (1) it never sums metrics across channels — a LinkedIn impression and a TikTok view are different units, so every comparison is within one channel; (2) it SUPPRESSES a verdict below 5 measured posts and says so, because a confident recommendation from 3 posts is worse than none; (3) a post with no recorded hook (published outside Hermoso, or backfilled without a creation match) counts toward channel and format totals but never votes on which hook works. Present the `finding` verbatim if there is one, and the reason if there is not. Read-only, 0 credits.

RECAST A VIRAL HOOK with the brand's own subject (the egg rolling off the table becomes your product; your host does the viral move), then join it into your clip. voice/language: a new voice, or language lip-synced, for one person swapped in. hook: a public post link (TikTok, Reel, Facebook, X; one lookup; not YouTube, which cannot be downloaded) or a video file. Subject: creator (saved name or id), image (a product or person; a real person's photo records consent), or neither for the brand's first product photo. A DUO OR GROUP (2-4 people): subjects [{creator or image, position: 'left'|'right'|'center' or your own words, refs?, outfit?}], each bound to its place in the frame, keeping that performer's movement and timing on the original sound. LOCATION AND OUTFITS, same pass: location (a place photo or words) moves the scene; outfit (a photo or words) re-dresses the subject, subjects[].outfit each person; with no subject named nobody is swapped and only those change (useBrand:true also swaps in the brand product). Pictures are references (4 max; more in words). Person for person: Seedance 2.5 (720p) if the hook is our render or faceRoute is set, else Wan 3.0 (≤15s; Kling 4.0 Edit past). FULL LENGTH: past its one pass (30 s) the window is recast in passes, each continuing from the previous one's frame and crossfaded across the cut, under the original sound: one video up to 60 s, quoted as passes plus join. mode 'auto' is always 'swap' (a region edit: camera, set, light, timing, text and sound kept; a multi-shot hook goes shot by shot); 'motion' only if asked (your person moves in their photo's setting). swap: your own words for what changes. start/end pick the window (3 s minimum; default: a group or a location/outfit change takes the whole clip, one subject the first 10 s). PAID, QUOTED FIRST: no confirm returns the plan and credits; confirm:true once the user okays. Then get_job until done, and post_edit with videoUrl = the result and ops [{op:'join', clips:[{url: your clip}], bridge:{kind:'impact'}}].

Copy the motion of any video (motion transfer): re-perform a clip's motion, camera and timing with YOUR character, product, clothes or place; sound kept. image (+ images, @Image1, @Image2… in prompt) = who and what. engine auto (default): a person clip + ONE photo of a person (saved, real or AI), 4-30s -> seedance (Seedance 2.5 through our face library, 720p: the clip's scene, beats and sound kept, the WHOLE person swapped) when the clip is one of our renders or faceRoute is set; otherwise wan (Wan 3.0, scene kept, 3-15s, 720p), past 15s kling-edit (Kling 4.0 Edit, 3-30s, 1080p by asking); several photos (one per person) -> h3 (H3 Max Recast, ≤4 photos, 5-30s); no person in the clip -> seedance. By name: kling-edit; wan = Wan 3.0 (≤15s, keeps the set); wan-prime; h3; kling = Kling Motion Control (ONE picture, its background becomes the set). voice/language: the new performer gets a new voice (same timing) or speaks another language (lip-synced), in the same job and price. Billed per output second, ~5 min per 5s; dryRun quotes.

Read this account's app settings — the LANGUAGE Hermoso writes ads and copy in (chat replies follow the language the user writes in), the app appearance (theme), and whether the weekly competitor-watch email is on. Same settings as the web app's Settings pane. Read-only, free.

Change this account's app settings. language = the language EVERY ad, script and plan is written in from now on (chat replies follow the language the user writes in) (say the language in plain English, e.g. "German", "Japanese", "Brazilian Portuguese") — it applies to renders made over MCP as well as in the app. theme = the app's appearance, "dark" or "light". watchEmail = the weekly competitor-watch email on/off. Only pass what you are changing. Account-wide (every profile), and it takes effect on the next call.

Delete one of the workspace’s CUSTOM skills by id (from list_skills). Built-in skills/recipes can’t be deleted. Minor + re-creatable, so no confirm needed.

Load a bundled skill’s full SKILL.md workflow instructions by name (from list_skills), or a built-in in-app skill by its id (e.g. youtube_script, retention_skeleton). Follow the loaded instructions to run that workflow with the other tools. Read-only, free.

Organic TikTok keyword search (there is NO TikTok ad library) — top-performing videos to mine for hooks/trends/remixable creative. Returns compact JSON {desc, author, handle, plays, likes, link, cover} per video, ranked by plays. Use research_ads for open-ended research; for what to POST, tiktok_ideas. Spends about a credit.

What to post on TikTok, from posts that are working THIS MONTH in the brand's niche: up to 6 content ideas (hook, format, caption, why, an act-within window of 7d or 30d, a quality tier), EACH backed by linked TikTok posts with their views, likes and engagement. One bounded pull of the last 30 days: up to 4 keyword searches + up to 5 hashtag pulls, then up to 8 reads of what the strongest posts SAY in their first seconds; only posts past 50,000 views back an idea, and an idea the catalog cannot back is dropped. Works with nothing connected. PAID and priced first: dryRun:true returns the exact credits (lookups + writing) with nothing run; the run charges exactly that (the writing part is not charged if no post clears the bar). THE PICK: call again with pick (one idea object from `ideas`) and to ('render_ad' or 'make_series'), free: it returns the next call's arguments (plan_ad then render_ad, or make_series) with the CLAIMS the ad may make, taken only from the brand's own facts. Nothing renders until you make that call. Omit keywords/hashtags to derive them from product/category or the saved brand.

Check a finished video for named, mechanical issues, ONLY when the user asks (never after every render, and it never re-renders): loudness vs a target (default -14 LUFS) and true peak, black and frozen stretches, captions or text outside the platform-safe band, text running off the frame. Optional assertions about named elements: appearsBy {element, bySec}, before {a, b}, staysInFrame {element}, keepsMoving {maxStaticSec}. Each issue comes with a Fix line. The loudness, black and frozen checks are free; the on-screen text and element read is a small paid frame read (frames:false skips it).

LOOK at a video's exact frames, free: one contact-sheet image of the frames from start to end at fps (e.g. a seam, 3.0-3.8 s at 10 fps) or at the listed times, tiles in reading order with each tile's second listed. Use it to find a moment before an edit and to check a render's seam after it. Takes a Library / render URL, a direct file or a public post link (a TikTok / Reel / Facebook / X post link is opened through one paid lookup, 1 credit; a file is free). SAVE STILLS: save:true returns full-size still images (frame grabs, up to 8) as URLs instead of a contact sheet, picked by times or by clips (numbered at the video's hard cuts, -1 = the last): a before / after from someone's own footage, a cover, a product shot. Compose with those URLs (make_template_ad template 'custom', a post) with no AI.

A video ad's structure: verbatim transcript (voiceover + on-screen text), beat list, duration and sampled frame times; study a reference before remixing it. ~A transcription call.

Pull an APP brand's REAL store screenshots (Google Play or App Store) into the workspace brand, so a screen-hungry native format can use them. Use when the brand has 0–1 app screens on file and you want make_template_ad(template:'app-ui-tour'), or the user asks to 'pull my app's screenshots'. A store link or package id goes in storeUrl and is read exactly; else pass appName (defaults to the saved brand's name). A web app with no store listing gets the screenshots of the product its OWN website shows (pictures the site labels as its interface). Call it before telling the user you have no screenshots of their app. FREE — a keyless store lookup or one read of the brand's homepage. It needs a CONFIDENT match: an ambiguous or unknown app returns 0 screens and saves nothing, which you should relay plainly rather than retrying with guesses. On success the screens are saved to the brand (durable URLs) and are immediately usable.

Resolve a generated asset reference (a /generated/… path or any URL) to a clickable absolute URL + a direct download URL.

Turn automatic credit reloads on or off (admin only): when the balance drops below a threshold, the card on file is charged for a top-up pack — SERVER-SIDE, even with no app open. Requires a saved card, added once in the app at first checkout/top-up; if there's none the tool tells you exactly where to add it. After that one-time card setup, agents can manage auto-reload, top-ups and plan links fully. Members (read-only billing) get an 'ask an admin' message.

Import this brand's PAST posts from a channel into the performance record, so 'which hook works' can draw on history rather than only on what was published since Hermoso started recording. Supports facebook, instagram, threads, youtube, tiktok, pinterest and bluesky; the others say plainly why they cannot (LinkedIn has no enumerate-my-posts endpoint on our grant, X bills per read so it is excluded from bulk import, Google Business has had no per-post insights since 2023, and Telegram's Bot API cannot read a chat's past messages at all — nothing published before Hermoso is recoverable through a bot token). BOUNDED, RESUMABLE AND QUOTED: it runs as a DRY RUN by default and tells you how many posts it found and what reading them will cost — pass confirm:true to import, and pass the returned cursor to continue. AN IMPORTED POST IS WEAKER EVIDENCE THAN A RECORDED ONE and is labelled 'backfilled': its hook is recovered ONLY where the post matches a Hermoso creation by asset or caption. A post made outside Hermoso stays UNATTRIBUTED — it counts toward channel and format totals but never votes on which hook works. Never guess a hook from a caption. Free.

Show this account's billing at a glance: current plan (quote plan.priceUsd per plan.period), credit balance, whether auto-reload is on, whether a card is on file, and whether YOU (this key) have ADMIN rights to change billing. Read-only, free. Call it before upgrade_plan / set_auto_reload to know what's possible — members have read-only billing. IN A SHARED TEAM WORKSPACE A MEMBER SEES THE PLAN AND THE BALANCE ONLY: the workspace owner's payment card and auto-reload belong to them and are not reported (billingScope:'member'). Never tell a member there is no card on file — the honest answer is that you cannot see it. plan.memberRate: media models at cost + 10%; plan.legacy: an older plan kept at its price. posting: this month's posting allowance ("N of 10 posts used this month" on Free, unlimited on every paid plan).

List every profile on this account (id + name; a profile is a brand, a creator or a personal workspace) and which one this connection acts on, PLUS any profile another account shared with you. Call this, then use_brand to switch; a SHARED profile switches the same way (pass its name or the profile id printed here). If a profile looks empty when the app shows it full, you are acting on a different one: call this first. Read-only, free.

Out of credits? Top up with a credit PACK. Call with no argument to list the available packs (id · credits · price). If the account has a saved card and you have billing-admin rights, calling with `pack` quotes the exact charge and calling again with confirm:true AND the quote's quote_token charges the saved card instantly (same one-click top-up as the app — no redirect). If there's no saved card yet, you get a Stripe checkout URL to hand your human for the FIRST purchase; their card saves for one-click after that. Packs only; subscriptions are managed by a person in Billing. IF YOU ARE AN AGENT HOLDING YOUR OWN PAYMENT CREDENTIAL, there is a third path that needs no human at all: `POST /api/billing/machine-payment` with a `packId` answers HTTP 402 carrying an MPP challenge, and grants the pack once you authorise and retry with the credential — the same packs, the same prices, the same credits. `GET /api/billing/config` carries a `machinePayments` block listing the packs with their per-credit rates and saying whether that lane is enabled on this server. Most agents do NOT have their own credential yet, so the checkout link above remains the normal path. To stop running out entirely, turn on auto-reload with set_auto_reload (admin) — low balances then top themselves up from the saved card automatically.

Run ANY Hermoso tool by name — including the paid-campaign, analytics and channel-admin tools that are not in this session's starting list — with the same permissions, the same account and the same result as calling it directly. Get the exact `name` and its `args` from find_tools first. This is the route on hosts that cannot reload their tool list mid-conversation (claude.ai, ChatGPT): enable_tools switches a group on server-side, but such a host keeps the list it fetched at connect time. Arguments are validated against the tool's own schema and a mistake is answered with the expected parameters, not a silent default. Refused by name, with the way out, when the tool needs a connector this workspace has not made or is withheld by the host's own policy.

Swap the narration of a finished video into a different voice — keeps the performance, lip-sync, and background sound. Use when the user likes the video but wants a different narrator voice; use dub_video only for language translation. Paid; returns the served URL.

Empty a range of cells in a Google Sheet, leaving the rows themselves in place. DESTRUCTIVE: call it WITHOUT confirm first and nothing is cleared — you get back the real number of filled cells in that exact range. Show the user that number, get an unambiguous yes, then call again with confirm:true AND confirmCells set to it. The echo is not ceremony: it is what catches naming A1:Z1000 when you meant A1:Z10, which is the mistake that actually happens. There is deliberately NO default range. The clear is read back and reported as confirmed only if the range really is empty afterwards. To remove a whole tab instead, use manage_sheet_tabs.

Paste a product page link (Shopify, Amazon or any store page) and get a creator-style video ad planned for that product. Hermoso reads the page (name, photos, what it says about the product, price), saves the product to the brand's catalog, checks the page's real product photo and pins it to this ad (the brand's saved default photo is not changed), then plans the ad. A store that hides its page from servers is read in a real browser; failing that the name comes from the link and the photo from a search, and the reply says so. Nothing renders here: show the user the concept, then call render_ad with the returned creative (dryRun:true first for the exact price). Works with or without a saved brand.

Cut ONE long video into several RANKED, ready-to-post short clips (podcast, webinar, interview, talk, long ad cut -> Reels/Shorts/TikTok). Transcribes with timestamps, picks the strongest SELF-CONTAINED moments, then cuts + reframes each with ffmpeg — no video model renders anything, so it is fast and cheap. THE VERTICAL REFRAME IS SUBJECT-AWARE: one cheap vision call per clip (billed as its own event) picks a SINGLE crop offset held for the whole clip, so a speaker sitting camera-left is not cropped out and the framing never drifts inside a clip; with nothing to discard or no single subject it stays dead centre — read `reframedToSubject` and each clip's `reframeWhy` back rather than assuming either way. ACCEPTS: a Vimeo / Loom / Dailymotion / Streamable / Rumble / Wistia / Twitch / TED watch link (NOT YouTube: YouTube does not let us download videos, so for a YouTube video the user downloads it and uploads the file), a direct https .mp4/.mov/.webm, or a Hermoso /generated/ URL (upload_file turns a local file into one). NOT supported: TikTok / Instagram / Facebook links, and anything age-restricted, private, members-only, geo-blocked or still LIVE — those fail fast with the real reason and are fully refunded, so ask for a direct file or an upload rather than retrying. Source ~15s to ~600MB; only the first ~40 minutes is analysed (truncated:true says so). Cost: a ~7-credit hold settled to the exact transcription + encode cost, plus the clip-selection model's tokens as their own small event. RETURNS clips[] — each its OWN served mp4 URL, title, hook, ready-to-post caption, 0-100 score and source timecode. SUBTITLES ARE BURNED IN BY DEFAULT (slim white CAPS, thin black outline, bottom safe band, no box) because short-form is watched on mute; captions:false for clean footage. TIMING IS APPROXIMATE (per-sentence, not word-level); never promise frame-accurate sync. captionsBurned counts the clips that really carry a burned track and captionNote says why any are bare.

Fetch fresh performance numbers for this brand's recorded posts and store them as a time-series. Metrics ACCRUE, so a post is read at ~24 hours and again at ~7 days; this collects whichever readings are due and skips the ones already taken. A channel that cannot report a metric records it as ABSENT with the reason — never as zero — and a read that fails is recorded as 'could not tell', which contributes to nothing. X IS SKIPPED BY DEFAULT because X bills us per API call: pass includeMetered:true to include it, and tell the user it costs credits BEFORE you do. The skip is always reported so a channel missing from the numbers is never mistaken for one that performed badly. Free except for X.

Discover a brand's competitor / similar / adjacent brands from its domain (Claude grounded by web search). mode=competitors (default, excludes the searched company), inspiration (best relevant ads incl. it), or company. Bills the discovery model's token cost (no ad-data charge); free inside a new account's first-brand setup.

Connect an account that links with a KEY rather than a sign-in screen, from here, with no browser: AppLovin Ads, Stripe, Intercom, Mailchimp, Klaviyo, Vercel, ChatGPT Ads, Apple Ads, Bluesky, Telegram, Bing Webmaster Tools, PostHog, Mixpanel, Amplitude, WordPress, Slack, Discord, Webhook. Pass provider and fields in that provider's own field names (listed below, * = required). The key is checked live with the provider before anything is saved, exactly as the app's Connectors page checks it, and the reply is read back from the saved connection. OFFER BOTH WAYS AND LET THE USER CHOOSE: a key pasted into this chat stays in this conversation's history, while pasting it in the app (Workspace > Connectors, or the one-click link https://app.hermoso.ai/?connect=<provider>) keeps it out of the chat. Hermoso never repeats a submitted key back. An account that connects through the provider's own sign-in screen (OAuth) cannot be connected here: this answers with the link to hand the user instead. Apple Ads with no key material first generates a signing key pair and returns the public key to register with Apple plus a setupToken to send back. Fields: applovin_ads {reportKey, eventKey, apiKey, capiKey, accountId} · stripe {apiKey*} · intercom {accessToken*, region} · mailchimp {apiKey*, server} · klaviyo {apiKey*} · vercel {token*, team} · openai_ads {apiKey*} · apple_ads {clientId, teamId, keyId, privateKey, setupToken, orgId} · bluesky {identifier*, appPassword*, pds} · telegram {token*} · bing_webmaster {apiKey*} · posthog {apiKey*, region, host, projectId} · mixpanel {username*, secret*, projectId*, region, workspaceId} · amplitude {apiKey*, secretKey*, region, host} · wordpress {siteUrl*, username*, appPassword*} · slack {webhookUrl, announceWebhookUrl} · discord {webhookUrl, announceWebhookUrl} · webhook {webhookUrl*}.

List the third-party accounts connected to this workspace (Meta, Google Ads, Google Drive/Sheets/Docs, YouTube, LinkedIn, OneDrive, Slack, …) — provider, status and the connected account label — PLUS which providers are available to connect. IT ALSO FLAGS A CONNECTION WHOSE PERMISSIONS ARE OUT OF DATE: a provider writes its granted permission set into the token at consent time, so a connection authorized before a permission was approved does not carry it and never will — those calls are refused by the provider and no retry or wait can change it. Check this FIRST when a connected provider starts refusing things. The fix is a browser: the user reconnects under Workspace > Connectors. Read-only, free.

Turn a file already in the user’s OneDrive into a PDF or a JPG — Microsoft does the conversion on its own servers, so nothing is re-encoded here and nothing is lost in a screenshot. It reads about 130 source formats, which is the point: PowerPoint and Word decks, Excel, Photoshop PSD, Illustrator AI, Sketch, 3D (fbx/glb/obj), video (mp4/mov/webm), HEIC from an iPhone, and the raw camera formats (CR2, NEF, ARW, DNG) that nothing else in this product can open. Use it to turn a client’s deck into images you can actually put in an ad, to get a usable JPG out of a designer’s PSD or a photographer’s raw file, or to hand someone a PDF of a spreadsheet. CONVERTING TO JPG REQUIRES BOTH width AND height — Microsoft refuses the call without them. The result is stored at a durable Hermoso URL you can pass straight to a render or a post; Microsoft’s own conversion link expires within minutes, so do not hand that one to anyone. Needs OneDrive connected — no new permission.

Remove a saved creator from this workspace’s cast by id (from list_creators). Records a cross-device delete so they don’t reappear on the user’s other devices. It only drops the roster entry — ads already rendered with that person are untouched — and the same portrait can be saved again with save_creator, so no confirm is needed.

WHAT TO FIX NEXT, post by post — and the tool that fixes it. post_performance tells you which hook is AHEAD; this tells you what is WRONG with a given post and where the next edit goes. Under-distributed is A HOOK PROBLEM (change the opening: mine_angles, then list_hooks, then plan_variations). Seen but not held is A RETENTION PROBLEM (plan_variations to rebuild the middle against the same hook). Seen, held, and still not converting is AN OFFER PROBLEM. FOUR REFUSALS, AND YOU SHOULD REPEAT THEM RATHER THAN PAPER OVER THEM: (1) a post younger than ~24h is TOO EARLY and is never called a failure — it has not had its run; (2) a metric the platform does not publish is UNMEASURED, never zero — Facebook has published no post reach since 2026-06-15, Reddit publishes no impressions, and Google Business publishes nothing per-post at all; (3) below 5 measured posts on a channel there is no baseline of the brand's own, and the ONLY fallback is a published short-video hook floor that is NOT our measured number and does not transfer off TikTok/Instagram/YouTube — it is attributed in the output and you should attribute it too; (4) it does not always find a problem, and 'nothing here needs fixing' is a real answer rather than a failure to look. Hermoso cannot see conversions for an organic post — no channel reports installs or purchases against a post id — so the offer rung runs ONLY when the user tells you they are not converting and you pass converting:false. Print `summary` verbatim. Read-only, 0 credits.

Disconnect a third-party account from this workspace (Meta, Google Ads, Google Drive/Sheets/Docs, YouTube, TikTok, LinkedIn, X, Reddit Ads, Pinterest, Google Business, Microsoft Advertising/OneDrive, Slack, …). This always drops the stored credentials, so every tool for that provider stops working immediately and posts/campaigns already published are NOT affected. IT NEVER REVOKES THE GRANT AT THE PROVIDER: the authorisation stays in place, so the same account keeps working in any other brand it is connected to, until the user removes Hermoso in that provider's own settings. The unconfirmed call names where to do that (list_connectors carries revokesAtProvider:"local") — relay it rather than promising a revoke. RECONNECTING NEEDS A BROWSER (the provider's consent screen) — an agent cannot undo this. Name the provider to the user, then call with confirm:true. Use list_connectors for the exact provider ids.

Fill in a profile (a business brand or a creator) from a website domain, a free-text description, or a social handle, into a {name, products, logo, …} object you can pass to plan_ad / generate. IMPORTANT: a domain can resolve to a DIFFERENT company than intended (e.g. bala.com is an engineering firm, not the Bala fitness brand at shopbala.com). Before spending any credits on research or renders, VERIFY the returned `name` (and `summary`) match the brand the user meant; if it looks wrong, re-draft with the correct domain or a description (pass save:false until confirmed) — this tool cannot ask the user, so the caller owns that check.

Localize a finished video into another language WITHOUT re-rendering it: each spoken line is said where it was said, in a native voice, lip-synced, background kept. Pass the video and the language; the script is read off the source (`script` only overrides it). Paid by length; dryRun quotes. Returns the localized video URL.

LIST a group of tools that is not in this session's roster. IT IS NOT HOW YOU REACH A TOOL — call_tool runs any Hermoso tool whether or not it is listed, and that works everywhere. Use this when the session will use MANY tools from one area and you want them in your list. WORKS ON CLIENTS THAT RE-READ THE TOOL LIST (stdio, the CLI, Cursor, Claude Code); a host that fixed its roster at connect time — claude.ai and ChatGPT do — will not show the new tools until it reconnects, and this tool says so in its reply rather than reporting a success you cannot use. The connect-time route that always works is `?tools=all` on the server URL. The default roster is CORE-FIRST: the core tools plus a few that make the connection drivable. Every other tool is held out of the LIST on SIZE alone — the whole registry is several hundred thousand tokens of schema re-sent on every turn, and a roster far past the 30-50 tool mark measurably degrades tool choice. The heaviest groups are `ads`, `analytics`, `channel_admin`: paid-campaign management across every ad platform is most of the total schema weight. NOTHING held out is unfinished or unsafe, and nothing is unreachable — find_tools finds it and call_tool runs it. CALL THIS WHEN A WHOLE AREA IS IN PLAY. If the user settles into building, budgeting, targeting or reporting on ad campaigns, call enable_tools({groups:['ads']}) and the tools appear. If they ask about their own site or product analytics, a tag/tracking container, or how a search engine crawls, indexes or ranks their site, call enable_tools({groups:['analytics']}). Groups: core, research, create, channels, channel_admin, analytics, ads, files, workspace — or 'all'. Free, instant, and it never turns anything off.

One error group in full by fingerprint (from list_errors): every field, plus the most recent redacted occurrences — status, connector, job id, workspace, and a shape-only echo of the inputs. This is what makes a bug reproducible. Read-only, 0 credits.

The errors actually recorded against this workspace, GROUPED by fingerprint — the same failure at the same call site is one row with a hit count and first/last seen, sorted defects-first. Each row says whose side it is: `ours` (a defect worth fixing), `user` (a refusal we deliberately authored, e.g. not-connected or out-of-credits), or `unknown` (a vendor 4xx we cannot attribute — never guessed). Free text, tokens, emails and creative are redacted before anything is stored, so an input echo shows shapes and lengths, not content. Filter by surface (http/mcp/agent/job/client) or kind. Read-only, 0 credits. Scoped to your own workspace; an operator whose client carries the admin key sees every account.

Before any paid render: does a picture or video contain a face, and which video models refuse it (they cannot take a real person's face; a render there fails after it starts) or were measured to take one, plus the model a face-bearing render is routed to. A local detector, no model call, 0 credits. It cannot tell a photo from a drawing, so a drawn face counts. Takes a picture or video URL, a Library item or a public post link. A file or Library item is free; a TikTok / Reel / Facebook / X post link is opened through one paid lookup (1 credit), refused when the account has no credits.

Surgically re-render ONE time window (1.5-8s) of an existing rendered video and splice it back on the VIDEO TRACK ONLY — the rest of the video and ALL audio stay byte-identical. Use when one beat/shot is broken ('the shot at 8 seconds glitches') and a full re-render would waste the parts that worked; bills only the replacement clip's seconds (~1/3 of a full render). Do NOT pick a window covering spoken dialogue (a video-only splice under speech breaks lip-sync) — pass speechWindows to enforce this. A FINISHED VIDEO-BLOCKS FILM (make_explainer's explainer or faceless channel; get_job says it is editable) is edited ONE BLOCK at a time instead: pass `job` (its job id) and `block` (1-based) or `atSeconds`, with `prompt` for a different shot and/or `line` for new narration words. Only that block is re-rendered and/or re-read; every other block, take, the music and the captions are the film's own and stay aligned. It is priced as that block alone and charged exactly what it quotes (dryRun). The original film is kept; a new film comes back, and it can be edited again. Feedback maps to one named move, never a re-render: quieter/louder (music db, audio_gain), faster/slower (speed; slower = zooms and pans at 0.7x), tighten (trim), retime (to a word anchor), recut (order, in-out points), regrade (timeline exposure/contrast/saturation keys).

Delete a saved Memory item by its id (from list_memory). Records a cross-device delete so it doesn’t come back. Minor + re-creatable (you can remember it again), so no confirm needed.

Make an exported sheet readable: bold the header row, FREEZE it so it stays visible while scrolling, and auto-size the columns so nothing is cut off. Worth calling right after create_sheet — a raw export with unsized columns and a header that scrolls away is the difference between a spreadsheet someone reads and one they close. Changes no cell VALUE, so it is never gated. The defaults do all three on the first tab; pass `tab` to pick another, freezeRows:0 to skip freezing, boldHeader:false or autoResize:false to skip those.

Structured Google Ads Transparency pull for ONE advertiser (by domain or advertiserId) — use when you know the brand; use research_ads for open-ended research. Deliberately fetches the cheap BASIC listing (get_ad_details=false, ~1 credit — the detailed variant with per-ad headlines costs 25 credits/call and is not exposed here). Returns compact JSON {advertiser, format, adUrl, image, firstShown, lastShown} per ad.

Turn ONE finished static ad into several copies that differ ONLY in the headline, for an A/B test: same picture, product, layout, colours and every other line. Pass `image`, and either `headlines` (your own, up to 10) or `count` (default 5, max 10) to have distinct angles written for you in the saved brand's voice (never inventing numbers, prices, ratings or claims the ad or brand does not state); `brief` steers what to test. The ad's text is read first (3 credits), then one image edit per headline; each output is proofread and flagged (textCheck) if the rendered words do not match, never silently re-rendered. When the saved brand has a product photo and the ad shows that product, the photo rides with the edit and the product's label is read and checked against it: re-printed from the photo only when it came out wrong (the check alone is charged when it is right; one small extra charge finds the product first). fixLabel:false turns that off. The whole batch is priced before anything runs. Returns each headline, its angle and its image URL.

HOOK MULTIPLIER: give ONE finished video ad N NEW OPENING HOOKS, N complete versions to A/B test. A planned hook replaces the first ~1.5-4 s (ends at the source's first cut there, else 3 s; hookSeconds overrides) with a new silent shot on a DIFFERENT named mechanic (list_hooks); the rest of the footage and the WHOLE original soundtrack stay, so every version keeps the source's length, and nobody in the hook talks or shows text. hooks[] adds your own openings: {url} a FOUND viral hook (post link or file) joined in front with the approved bridge (cut before its payoff, its own payoff sound carried across; post_edit price), {prompt} an opening described in words, {mechanic} a named one; with hooks and no count only those are made. To make the found hook's subject your product or creator first: recast_hook. Pass the video's FILE URL (a render, job result, list_library or upload_file). 1-5 versions, default 3. Refused free before billing: a source over 120 s, too short for a 1.5 s hook plus 2 s after it, unreadable, or a social post as the SOURCE (clone_video remakes someone else's ad). COST: quoteOnly:true returns the exact credits free: state that first. A paid planning read (planCredits) runs next, then each planned version bills like fix_beat for the hook's seconds. dryRun:true pays for that read; pass its `plan` back to render exactly those. Returns ONE JOB PER VERSION; call get_job on each until done, never describe a version before its URL arrives. Hooks that show the product use the brand's product photo (productImage overrides; useBrand:false sends none).

The curated menu of VISUAL scroll-stop HOOKS (how an ad opens) and SETTINGS (where it is staged) that plan_ad and render_ad accept, PLUS this brand's measured traction per hook. Call it before planning an ad to pick a hook deliberately instead of letting the model improvise one, and call it after publishing to see which ones are actually landing. Three things it will not do: it never recommends a hook from thin data — a verdict is SUPPRESSED below 5 measured posts and the reason is stated; it never compares across channels; and it reports hooks you have NEVER TRIED as a fact, not as advice, because 'you haven't tried this' is an observation and 'you should' would be a verdict drawn from zero data. A hook marked unusable in this brief says WHY (an on-screen-text hook cannot ride an authentic/UGC render, which carries zero on-screen text). Read-only, 0 credits.

Pull the files in a Google Drive or OneDrive FOLDER into this profile's Library, so they can be used like anything rendered here — published, scheduled, cloned, used as a product photo or a reference. Hermoso downloads each file with the user's own connected account (a Drive/OneDrive file is not public) and stores a durable Hermoso url for each. Give `folderId` from list_drive_files / list_onedrive_files with onlyFolders — omit it for the root. GOOGLE DRIVE ONLY SHOWS WHAT THE USER HANDED OVER: our Drive scope is `drive.file`, so Hermoso can see the files and folders it created plus the ones the user picked with the Google picker in the app, and NEVER their whole Drive. Picking a FOLDER hands over the folder alone, not the files in it, so if a folder comes back empty the reply carries a link that opens the picker inside it, where the user selects every file at once (Cmd+A / Ctrl+A); give them that link. SUBFOLDERS ARE NOT WALKED and Google-native docs (Docs/Sheets/Slides) have no file to download: both are reported back BY NAME rather than silently dropped, along with anything too large or unreadable, so you can tell the user exactly what did and did not come across.

INSERT a newly generated scene into a clip you already have: the video plays up to startTime, the new scene (5 to 13 seconds) plays, then the original picks up again at resumeTime. Use for a product close-up, a reaction, a reveal or a missing beat in the middle of an ad without re-rendering the rest. resumeTime = startTime keeps every second of the original; a later resumeTime also cuts the footage in between. Describe the scene in prompt and/or attach up to 4 reference pictures (refImages) or 1 reference clip (refVideos). Clips up to 60 seconds; both times between 1.625 and 60 seconds. Paid render (billed per second of NEW scene); returns the whole video's URL.

Organic Instagram REELS keyword search (/v2/instagram/reels/search — our only IG keyword surface; profile/hashtag pulls go through fetch_social_data with a handle). Returns compact JSON {desc, author, handle, plays, likes, link, cover} per reel, ranked by plays. It matches captions worldwide and cannot be limited to a country or language: for one market, search in the words people there use, or read named local accounts' Reels through fetch_social_data's Instagram profile endpoint, which is the better first step for a local market. Spends about a credit.

Invite someone to this profile by email (role: member = read-only on billing, admin = full). This SENDS a real email invite / share link — an account change. Confirm the exact email + role with the user, then call with confirm:true.

Poll a render job by id. Returns status (queued|running|done|error), progress, and on done the served media URL. Renders take 1–3 minutes: keep calling this until done/error without asking the user — several calls is normal, not a stall. An id that does not exist on this account answers status "not_found" — that is FINAL: stop polling it, and do not re-fire the render (that double-charges).

List the most recent render jobs + how many are currently running, so you can report on or resume in-flight work.

On a connector that several teammates can contribute their OWN account to (see list_connectors — the row reports multiContributor), remove YOURS from this profile: your stored credential is dropped and the accounts you shared stop being shared. Your teammates' accounts on the same connection keep working, and nothing changes at the provider — reconnecting in a browser shares again. Use this instead of disconnect_connector when the connection is not yours to remove: disconnect_connector revokes the grant at the provider and only the person who created the connection may call it.

Structured LinkedIn Ad Library search by company name, keyword, or companyId — use for a targeted B2B pull; use research_ads for open-ended research. Returns compact JSON {advertiser, headline, description, cta, link, media, dates, impressions} per ad — LinkedIn's library includes impression counts. Spends about a credit.

Translate the on-image text of ONE finished static ad into other languages and keep everything else: same picture, layout, typeface, colours, logo and product. Pass `image` and `languages` (up to 5, e.g. ["Spanish", "German", "French (Canada)"]). The ad's text is read (3 credits), translated the way a native copywriter in each market would write it (brand and product names, URLs and prices kept as written), then one image edit per language; each output is proofread and flagged (textCheck) if the words do not match, never silently re-rendered. When the saved brand has a product photo and the ad shows that product, the photo rides with the edit and the product's label is read and checked against it: re-printed from the photo only when it came out wrong (the check alone is charged when it is right; one small extra charge finds the product first). fixLabel:false turns that off. Priced before it runs. For a VIDEO use dub_video.

Add, rename or delete a tab in a Google Spreadsheet. action:"add" + title · action:"rename" + tab + newTitle · action:"delete" + tab. Name the tab by its TITLE or its numeric sheetId (list_sheet_tabs gives both); an unknown tab is refused with the real list rather than a Google error nobody can map back. DELETING a tab destroys everything on it: call it without confirm first to get the filled-cell count, then confirm:true + confirmCells. Google does not allow removing the LAST remaining tab in a file, and that is refused by name with the way out (clear it, or delete the whole file with delete_drive_file). Every action is read back from the spreadsheet before it is reported as done.

Remove a member from this profile by email — they lose access (you can re-invite them later), AND every connection THEY made on this profile is disconnected with them: their own X, LinkedIn, TikTok, YouTube, Pinterest, Threads… connections (on a channel holding several accounts, only the accounts they added), and any extra login or authorization they added. Connections the owner or anyone else made are never touched, and one with no record of who connected it is treated as the owner's. Queued posts that would publish through their accounts, or that they scheduled themselves, will not go out. The unconfirmed call reports exactly which connections and how many scheduled posts — relay that to the user, then call with confirm:true.

List the durable facts & preferences saved in this workspace’s Memory (what the studio remembers about the brand, audience, taste, and do/don’t rules) — the same Memory the web app shows. These shape every future ad. Read-only, free.

Mine ad ANGLES from real customer language: gathers the customer's own words (Reddit, TikTok, the brand's review page + review-site results) and returns a RANKED angle bank — each angle tagged (pain / outcome / identity / fear / competitive-displacement / social-proof / contrast), 2-5 VERBATIM proof quotes, a 0-100 score with breakdown, and a ready-to-run hook in the customer's own voice. Reads YOUR saved brand (pass brandId to target a specific profile — that switches this key's active profile like use_brand). To tear down a COMPETITOR use competitor_teardown instead. YOUR OWN REVIEWS: pass `reviews` (a list of review texts, or one pasted block: one per line, numbered, blank-line separated, or a CSV with a review column) and/or `reviewsUrl` (a CSV, TXT or JSON file from upload_file, or a review page; on a local CLI a file path works too). They are first-class evidence: every quote from them is checked word for word against what you sent and labelled 'your reviews', and a quote that is not verbatim is dropped and counted. useOwnReviewsOnly:true mines only your reviews and gathers nothing public. Limits: 300 reviews, 2,000 characters each, 40,000 in total; over that it is refused at no cost, so send fewer or split into batches. Each angle comes back with `next`: the exact plan_variations and generate_image arguments that turn it into finished statics (one generate_image per angle = statics with distinct angles). Spends credits on the reads and the model's token cost.

MULTIPLY a winning video ad into N variants: each gets a NEW character, outfit, location and/or objects while the cut, the camera motion, the pacing and the ORIGINAL AUDIO stay exactly as they were, and any burned-in captions are removed. Pass the source video URL. HOW: the source video itself DRIVES each variant (motion transfer from one image-edited opening frame), so every variant keeps the source's length, cut and audio; only the person, outfit, set and props change. Sources up to 30 seconds work as they are; longer ones are refused for free with the way out (trim it first: post_edit with ops [{op:'trim', start:0, end:30}] — clip_video is the AI highlight clipper, not a trim). ONE JOB PER VARIANT: get_job each until done; never describe a variant before its URL arrives. COST: quoteOnly:true returns the exact credits free: state that first. A paid planning read (planCredits) runs next, then each variant bills perVariantCredits. DryRun:true pays for that read; pass its `plan` back to render without paying again. Images takes up to 9 of the user's product, people, clothes or places, and change names them @Image1, @Image2 in order ('she holds the bottle in @Image1'); the saved brand's product photo is read as a reference to its real look (useBrand:false leaves it out). A TikTok, Reel, Facebook or X post link works as the source (not YouTube: upload the file): one paid lookup reads it into a file. Quality: 1080p (default) or 720p; quoteOnly prices both. Regions: pass regions:['Berlin','Tokyo'] to restyle variants per market; translation is separate: dub_video on a finished variant. `change` is what should change, in the user's own words ('older women', 'winter streets', 'swap the mug for our bottle'): every variant applies it and the variants still differ around it; a part of it about the voice, words, language, music, length or captions cannot change here, and the reply says so and names the tool that can.

Delete a saved playbook by id (from list_playbooks). Records a cross-device delete so it does not come back on the next sync. Minor + re-creatable, so no confirm needed.

List the PLAYBOOKS saved in this workspace — the reusable strategy cards (winning hooks, angles, formats and the concrete plays to run) kept from teardowns, angle mining and creatives worth repeating. The same Playbooks the web app's Playbooks tab lists. Read one before planning an ad so you re-run what already worked instead of starting cold. Read-only, free.

Pre-flight ad copy against Meta's REAL, live Advertising Standards before you run it — a flat 1-credit check. Pulls Meta's actual policy pages and returns a verdict (pass / fix / block) where every flagged issue QUOTES Meta's own policy text verbatim plus a compliant rewrite that keeps the sell. It's a check, not an edit — it never changes the creative. Especially worth running for regulated-adjacent categories (health/supplements, weight-loss or beauty results claims, finance/crypto/insurance, alcohol, dating, gambling) or ANY strong/absolute/guaranteed claim.

The POSTS people make ABOUT a subject — a brand ("liquid death"), a product, a hobby ("coffee"), a hashtag ("#homecafe") — from whoever posted them, across organic TikTok, Instagram Reels and YouTube in ONE call, ranked by views. Not the brand's own ads (search_meta_ads / research_ads) and not the people (find_creators folds these same posts into creators): use it to see what is actually being posted and watched about a subject, to find clips worth cloning (clone_video), and to read the hooks and angles an audience already responds to. About one credit per platform searched (one query each by default; `queries` adds "best X" / "X review" / #tag variants, each a paid call); repeats inside 20 minutes are free.

Put a PRESENTER on ANY video, only when the user asks for one: their pick of person (`creator`: a saved or preset creator from list_creators; or `image`: a photo they gave; never invent one), filmed on a green screen saying the words, cut out and laid over the video in a bottom corner (default), a round badge, or split screen, lip-synced, block by block across the whole video (about 10 s and about 25 spoken words a block). The words are `script` (verbatim; refused free if it cannot fit) or, without one, written to fit from what the video says and shows (`brief` steers them). The video's own sound stays under the presenter unless baseAudio 'off'; the video without the presenter is kept. A face the video model refuses STOPS it with nothing charged: offer another creator, another photo, or the video as it is. dryRun:true = the exact price, which is what it charges. For a finished make_explainer film, make_explainer with fromFilm.

The brand's product catalog: each product's id, title, page link, price (and whether it is current), photos, its own selling points and audience, and when it was last checked against its live page. Free. `refresh:'stale'` re-reads rows older than a day first, `refresh:'all'` re-reads every page. Use the ids with make_static_ads.

List every post Hermoso has recorded publishing for this brand, newest first, across all channels — channel, permalink, caption, format, the HOOK and SUBJECT it was written to, and its measured engagement. The WHOLE history, no cap: pass the reply's nextCursor as `cursor` for older posts. Each row says how it was recorded: 'captured' (written at publish time — the hook is what the author intended) or 'backfilled' (reconstructed from the platform afterwards). A dash for engagement means the platform reported no number — NOT zero. Read-only, 0 credits.

Use this when the user asks which ads a named brand is running. Pull one named brand’s live ads from the Meta (Facebook and Instagram) Ad Library: deduplicated, sorted, with the brand’s own page resolved. One call, back in a few seconds. Use it when the user names a brand and asks what ads it is running ("show me the ads <brand> is running"). Covers the Meta Ad Library only; for a broader search across brands or platforms use research_ads. Spends credits.

Reddit keyword search (/v1/reddit/search, top-ranked) — a goldmine for the customer's OWN words (pain points, objections, language) to mine into ad hooks and copy. Returns compact JSON {desc (title+selftext), subreddit, upvotes, comments, link} per post. Spends about a credit.

Reframe a video to a different aspect ratio (e.g. 16:9 master -> 9:16 vertical): the whole original frame is kept and the new canvas around it is generated, with the sound and timing unchanged. Clips up to 300 seconds. Paid render; returns the served URL of the reframed video. Every placement at once (a free blur-pad by default): repurpose_video. A presenter on any video: add_presenter.

Re-read product pages now (all products with a page, or `ids`) and update the catalog: the live page wins, a failed read is recorded and said. `sync:true` instead imports the whole store catalog (the brand's connected Shopify store, else its website's public catalog), merging into existing rows. Free.

Save a durable fact or PREFERENCE about the brand, audience, or the user’s creative TASTE (e.g. “audience is first-time homebuyers”, “prefers bold lime accents”, “always captions off”) into the workspace Memory so it shapes FUTURE ads. For lasting things, not one-off requests. Merges into the existing Memory (never overwrites); de-dupes on identical text. NEVER how Hermoso, a tool, a connector or a platform API behaves (what a call returns, errors, permissions, limits, ids) and never a phone number or email — those are refused.

The OLD NAME of clone_static, kept so agents that already call it keep working. It is the same tool with the same inputs, result and cost; prefer clone_static.

Use this when the user asks for a video ad, Reel, TikTok ad, UGC video or commercial for their business or product It renders a finished ad on a real video model, on their brand (real product photos, brand voice), as an mp4 ready to post: timed shots, clean speech, text set in post, no music, captions or end card unless asked. Pass plan_ad’s full output as `creative` (plan_ad turns a one-line ask into the concept). UGC and cinematic plans render from a painted STORYBOARD (a cinematic spot paints a location anchor first, and a hero when someone is on camera); a product-only spot renders from the real product photo, and board:true or foundations:'choose' adds an identity sheet, a moodboard and a board. A plan that fits one clip of the render model renders as one continuous pass; a longer one as STITCHED ACTS (the fewest balanced clips), never time-compressed. Most models cap a clip at 15s and the longest at 30s: dryRun:true shows the act split free, before spending. The brand’s saved product photo (get_brand / list_product_photos) is the real product in the shots, with no upload needed. CAST A SAVED CREATOR with `creator` (list_creators) so the same person stars in every ad; otherwise each render casts a new face. The user’s own pictures (creative.inserts, or `inserts`) are laid in full screen at their seconds. An app shown on a device screen with no real screens to lay there returns needsAppScreens, nothing charged: ask, then pass screenChoice or screens. 1–3 min; poll get_job while it renders. Spends credits.

Report a bug in Hermoso to the team. Use this when something in Hermoso genuinely misbehaves — a tool errors unexpectedly, returns a wrong or malformed result, a render comes back broken, or documented behaviour doesn't match what happened. Include what you were trying to do, the exact tool call and arguments, and what came back. Do NOT use it for out-of-credits, a policy refusal, or a missing capability (use request_feature for that). Free, no credits.

Ask the Hermoso team for a capability that doesn't exist yet. Use this when you need something Hermoso genuinely can't do — an unsupported platform or channel, a missing model, an export format, a tool that would have completed the user's task but isn't available. Say what the user was trying to achieve, not just the feature name — the use case is what gets built. Free, no credits.

Use this when the user asks an open-ended ad question across several brands or platforms, such as comparisons, which angles are working in a market, or ad examples for a category. Do not use for one named brand’s current ads: pull_competitor_ads covers that. Open-ended ad research that needs JUDGMENT across platforms — comparisons, "what angle is working", "who else is doing this", anything where the right sources are not known up front. It is an agentic loop (several rounds of library pulls plus a written synthesis) and typically takes 30-60 seconds, so it is the WRONG tool for a question that names its own answer. For one named brand’s live ads use pull_competitor_ads; for one keyword or one advertiser on Meta use search_meta_ads — both are a single call and return in a few seconds. Spends credits — an agentic loop, so a handful rather than the one-call cost of a targeted search.

Re-lay out ONE finished static ad for other placements: the same ad, product, copy (word for word), logo and style, recomposed natively for each canvas rather than cropped. Pass `image` and optionally `aspectRatios` from 1:1, 4:5, 9:16, 16:9, 3:4, 4:3 (default 1:1, 4:5 and 9:16; the ad's own ratio is skipped). Reads the ad's text first (3 credits) so every line survives, then one image edit per canvas. When the saved brand has a product photo and the ad shows that product, the photo rides with the edit and the product's label is read and checked against it: re-printed from the photo only when it came out wrong (the check alone is charged when it is right; one small extra charge finds the product first). fixLabel:false turns that off. Priced before it runs. Several sizes AND versions (headlines, languages, offers) at once: repurpose_image. For VIDEO use reframe_video.

RESTYLE a clip into a new LOOK: every frame redrawn, shots, motion, camera, framing, timing and sound kept. Looks: `Restyle looks` in hermoso_capabilities (claymation, knitted yarn, cel-shaded CG anime…), your own words, or a look BUILT FROM YOUR IMAGES (styleImages); characters draws your own characters in. engine auto (default) picks by what is in the clip: a real-looking person -> kling (Kling 4.0 Edit, 3-15s); none -> wan (Wan 3.0, recommended, ≤15s) else seedance (Seedance 2.5, 4-30s); wan-prime = Wan 3.0 Prime by name (costs more). Paid; dryRun quotes it. One change only: edit_video.

Finish a video-blocks explainer or faceless-channel film (make_explainer) that STOPPED partway: a block the video model would not render, or a server restart. What it already made is kept (the script, the style key, the narration and every block that rendered), so this renders ONLY the missing blocks and re-assembles the same film, with the same topic, look, voice, length and options. It reserves credits only for what is missing and never charges again for what was kept. get_job and list_jobs show which jobs can be resumed and which blocks are kept. Quote it with dryRun first and tell the user the number. A finished film is not resumed: to redo one block of a finished film, use fix_beat with its job id.

Change a workspace member’s role — admin (full access incl. billing) or member (read-only on billing). A privilege change: confirm the exact person + new role with the user, then call with confirm:true.

Set the OUTREACH STATUS and/or a NOTE on a creator already saved in the swipefile (find_creators -> save_to_swipefile, or the heart on a creator card). Status is one of new | contacted | replied | booked | passed. The note is free text (deal terms, rate, what was sent). Reads back the updated row. Use list_swipefile to find the key. Free.

Virality/performance prediction for a finished ad (image or video URL): overall score, per-dimension breakdown (scroll-stop, hook, clarity, brand/product, CTA, retention, goal fit, own: could another brand run it unchanged), strengths, and the single biggest fix. Use BEFORE spending on distribution, or to rank variants.

Create a SHARING LINK to a OneDrive file or folder (fileId from list_onedrive_files): type view (read-only, default), edit, or embed (an <iframe> link, personal OneDrive only); scope anonymous (anyone with the link, default), organization (work or school OneDrive only) or users (only people you invite); optional expiresAt (date or ISO) and password (on a personal OneDrive both need a Microsoft 365 subscription; the refusal says so). An ANONYMOUS EDIT link lets anyone holding it change or delete the file, so it needs confirm:true after you show the user that; every other link is created straight away. If Hermoso already made the same kind of link, OneDrive returns that one and the reply says it was reused. Read back from the item's permission list. 0 credits. Needs OneDrive connected.

List the bundled Hermoso SKILLS — multi-step workflow instructions (SKILL.md) that orchestrate the other tools (research an ad space, plan+render a finished ad, product photoshoot, raw generation) — plus the in-app strategy skills and creative recipes. Call get_skill to load a bundle. Read-only, free.

Generic escape hatch for any ALLOWLISTED long-tail social/web endpoint the dedicated search_* tools don't cover — e.g. {path:'/v1/instagram/profile', params:{handle:'nike'}}. Allowlisted platform families: TikTok (+ TikTok Shop), Instagram, YouTube, Facebook (organic profiles/posts/events/marketplace), LinkedIn (organic posts/companies), Twitter/X, Reddit, Threads, Snapchat, Pinterest, Twitch, Bluesky, Truth Social, Rumble, Spotify, SoundCloud, GitHub, Google search, link-in-bio pages (Linktree etc.). Param names vary per endpoint (profiles use `handle`, keyword searches use `query`, Reddit uses `subreddit`). WARNING: returns RAW provider JSON — large and messy; prefer the dedicated search_* tools. Spends credits.

A sound for an edit by name ('the FAAAA sound'), by the moment ('bad news reaction') or by link (TikTok sound, meme-sound page, post, audio file): a durable mp3 and where it starts and lands. Named/described sounds come from what TikTok uses now; pick takes another candidate. A sound that does not exist yet is generate_sfx.

Render a multi-scene STITCHED video (≥2 scenes) — ONLY for spots LONGER than ONE clip of the chosen model. A multi-beat ad that FITS one clip renders better and cheaper as ONE single-pass generate_video/render_ad (a single generation carries the whole hook->demo->payoff arc) — never stitch those. What fits is the model’s own maximum from hermoso_capabilities, not a fixed number: 15s on most models, 30s on the longest-clip one, so a 30s spot need not be stitched at all if you name that model. Blocks until done. Spends credits.

Read one of this workspace’s data stores by key, for visibility into what the app holds — playbooks, swipefile, saved locations, avatars, creations, chats, brand, memory, skills. Read-only, free. Allowed keys: heist.memory.v1, heist.skills.v1, heist.playbooks.v1, heist.avatars.v1, heist.locations.v1, heist.chats.v1, heist.creations.v1, heist.assets.v1, heist.brand.v1, adInspo.swipefile.v1. (The typed tools — list_memory / list_skills / get_brand — are friendlier for those; use store_get for the rest.)

Burn subtitles into ANY video and get the .srt too; nothing is cut or re-rendered. Set textStyle only when the user describes a look (default: white sentence case, thin outline, bottom). Timed per sentence, or auto:'words' for phrases on their spoken words. Only when asked, textStyle 'highlight' (or a highlightColor) lights the word being said, timed word by word. Or pass cues to burn your own lines exactly, untranscribed (reel-style phrases, *stage directions*). burn:false returns only the .srt. Takes a video file or URL, a Vimeo-style page, or a TikTok / Reel / X post link (not YouTube: YouTube does not let us download videos, so upload the file). No speech is refused and refunded. Runs in the background and lands in the Library.

List this workspace's SWIPEFILE — the saved-ad research board: every named collection and the ads/creatives kept in it (advertiser, headline, body copy, media URL, platform, when it was saved, and any taste tags). The SAME board the web app's Swipefile tab shows. Use it to answer "what have we saved?", to mine the user's own taste before planning an ad, or to find a reference to remix. Read-only, free.

List the members of the current profile — email, role (admin/member) and status. Read-only, free.

Ready-to-run ad BRIEFS by industry: dropshipping, ecommerce, mobile apps, restaurants, real estate and local services (12 trades: plumbers, electricians, hvac, roofers, landscapers, house-cleaners, pest-control, auto-repair, salons, med-spas, dentists, gyms). Compact list by default; pass `id` for one template in full: its brief (filled with the active profile's name, and the trade's specifics when `trade` is given), what the user must supply, the recommended format/aspect/length, and its web deep link. To run one, fill the [square-bracket] slots from what the user gave you, then pass the brief to plan_ad as `product` (with its format and durationSeconds) and render the plan (render_ad for video, generate_image for images). Every part is editable: a template is a shortcut, not a lane, so change or ignore anything that does not fit. These are NOT the native HTML formats of make_template_ad. Read-only, 0 credits.

Organic Threads keyword search (/v1/threads/search) — short-form text/social posts for trend + voice research. Returns compact JSON {desc, author, handle, likes, link, cover} per post. Spends about a credit.

Clean up and consolidate the workspace Memory: drops entries that are about how Hermoso, a tool, a connector or a platform API behaves (product behaviour, not the brand), drops phone numbers and emails, and merges near-duplicate facts into one sentence each. Call with no argument to get the PROPOSAL (what would be removed and merged, with reasons) — nothing changes. Call again with confirm:true to apply it through the same typed writers the app uses (deletes carry tombstones so they stay deleted on every device). One model call, billed at its token cost.

Search EVERY Hermoso tool — your starting list is deliberately short, and everything else in the product is here — by name, task or group. Each row gives the tool's PARAMETERS in one line, its CREDIT COST (free means free on every plan; a tool that runs a model quotes the live per-model figure) and its recent HEALTH on this server (failure rate and typical duration, or "no recent calls", which means unseen and not broken). Use it the moment the user asks for something you do not see a tool for (a campaign, an ad set, a lead form, a click-to-WhatsApp ad, a report, keywords, audiences): a tool missing from your list is NEVER proof the feature is missing. Then run the tool with call_tool. A tool that is failing or needs a connector this workspace has not made is ranked last and marked, never hidden — pass onlyHealthy:true if you want those left out. Free, read-only.

Change the SUBSCRIPTION plan (admin only). No argument lists the monthly plans. Basic ($9.99) is the posting plan: unlimited posting and Autopilot with NO monthly credits (it includes the same 250+ free credits as the Free plan, monthly credits start on Starter, and Autopilot spends credits). Starter, Pro and Studio include monthly credits and no markup on models. Every paid plan posts without a limit (Free: 10 posts a month on 3 channels). New subscriber: a Checkout URL for the human. Existing subscriber upgrading: a quote + quote_token; once the human agrees, call again with confirm:true and quote_token to charge the card on file. Downgrade: a billing-page link; it starts at the term's end.

Persist an ARBITRARY user file (image, video, audio, PDF or document, up to 150MB) into Hermoso and get back a durable public URL that EVERY publish, schedule and ad-build tool accepts. Kept in the Library. THIS IS THE BRING-YOUR-OWN-CREATIVE PATH: it is for files that have NOTHING to do with a Hermoso render (media on the user's desktop, an agency's finished ad, a photo they shot), and it means you can publish, schedule and run ads through Hermoso without generating anything here. Provide exactly ONE source — passing two is an error, never a silent preference: `url` (ANY public http(s) link — Hermoso fetches it server-side, so nothing crosses this connection and there is no practical size limit; THIS IS THE ONE THAT ALWAYS WORKS, including on the hosted connector), `dataUri` (a base64 data: URI — keep it under ~15MB, since the bytes travel over this connection). If the file is already at a public https URL, the Meta, Reddit and ChatGPT-Ads tools take it directly and re-host it safely. When in doubt, use this: its URL works everywhere. (Calling the underlying HTTP route directly? POST /api/upload takes the file's RAW BYTES as the request body with its own content-type — NOT multipart/form-data — or ?url=<link> with no body.) Returns {url, kind, bytes}.

Upscale a video to higher resolution (2x) for final delivery. Paid render; returns the served URL. A 480p Seedance 2.5 draft: finish_draft (same take, native 1080p). Three engines: the default ('standard') is the safe precision upscaler; engine:'flux' is the FLUX 3 video upscaler (1080p/2K/4K) with an optional mode:'creative' detail-enhancement pass; engine:'detail' rebuilds fine detail such as hair, skin and fabric, the sharpest result for a hero shot and the most expensive. Pick 'flux' or 'detail' when the user asks for added detail rather than a faithful enlargement.

Pin which profile this connection acts on. Pass the profile id or exact name from list_brands. Works for your own profiles AND one another account SHARED with you (pass its name or the profile id list_brands prints); access is verified against your real membership before it is pinned. Persists for this API key until changed, on every surface (hosted connector included — no environment variables, no restart).

Read what a brand's MARKET RADAR (set_competitor_watch) has found: each watched brand's new ads, the winners (ads past 30 days running, with a ready brief to make your own version), the trending formats and top posts for its topics, and the ideas to make, plus its state (lastRun, nextRunAt, lastRunOk, paused) and the quote for its next check. Findings marked `seed:true` are what a brand was already running at its first check, not new launches. Read-only and free; it never runs a check.

Organic YouTube keyword search (/v1/youtube/search) — videos to mine for hooks/angles/long-form structure. Returns compact JSON {desc (title), author, handle, plays, link, cover} per video, ranked by views. Spends about a credit.