Higgsfield Workspaces

Higgsfield Workspaces — Task-First Decision Layer

Introduction

Higgsfield is organized around workspaces, each one tuned for a specific kind of

production work. The right starting question is not "which model should I use?"

but "what am I actually trying to make?" Inside a workspace, model choice still

matters — it shapes quality, speed, and cost — but a workspace mismatch costs

more time than a suboptimal model. Pick the workspace by the task, then pick the

model by the result you want.

The Decision Matrix

| Your production problem | Start with workspace |

|-------------------------|----------------------|

| Cinematic scene with deliberate camera direction | Cinema Studio |

| Speaking character, dubbing, or avatar video | Lipsync Studio |

| Rough idea, sketch, or storyboard test | Draw to Video / Sketch to Video |

| Fast short-form or viral-style content | Sora 2 Trends |

| Product ad or ecommerce variation | Click to Ad |

| Narration, voice swap, or translation | Higgsfield Audio |

If the user's task matches more than one row, pick the workspace whose output

most closely resembles the final deliverable — you can always chain into another

workspace for polish or audio later.

Workspace Descriptions

Cinema Studio

The professional filmmaking environment on Higgsfield. Built for multi-shot,

character-consistent sequences where camera direction, genre framing, and visual

cohesion matter. Cinema Studio 2.5 offers optical physics (camera body + lens

stack) and built-in color grading; Cinema Studio 3.0 (Business/Team plan) adds

native audio, Smart camera planning, and longer durations. Choose this workspace

when the work needs to feel like a film. For prompt construction, Director Panel

use, optical stack selection, and multi-shot planning, see higgsfield-cinema.

Lipsync Studio

The workspace for putting words into a character's mouth. It pairs an image,

video, or generated face with an audio track and synchronizes the lip movement

to the speech. Use it for dubbed performances, talking-head avatars, character

dialogue, and lipsync work layered on top of existing video generations. For

audio layer design (dialogue, SFX, ambient, BGM) that feeds into Lipsync Studio,

see higgsfield-audio. For character reference management and Soul Cast avatar

generation that commonly sits upstream, see higgsfield-soul.

Draw to Video / Sketch to Video

The workspace for taking a rough sketch, blocking diagram, or storyboard panel

and turning it into a generated video. It exists to bridge the gap between

paper-stage ideation and motion testing — the place to validate a shot idea

before it earns the credits and time of a full Cinema Studio build.

When to use:

• Early ideation, when you're still deciding what the shot wants to be.
• Shot blocking experiments — testing how a composition reads in motion.
• Pre-production validation before committing to Cinema Studio scene work.
• Fast iteration on framing, staging, and silhouette before the prompt or

reference set is locked.

Input characteristics: the workspace is forgiving. Hand-drawn storyboards,

blocking diagrams with stick figures or geometric placeholders, even a crude

napkin sketch all work as input. Quality of line is not the point — clarity of

compositional intent is. A messy sketch with a clear staging idea outperforms

a polished sketch whose staging is muddled.

The prompt's role: the sketch carries composition and blocking; the prompt

carries everything else. Subject details, environment, lighting, mood, and

material qualities all live in the prompt. Think of the prompt as describing

the realized scene the sketch was hinting at — the sketch told the engine

*where* things sit and *how* they move; the prompt tells it *what* they are.

Output expectations: test-quality, short. The output is a tool for shot

validation, not usually the final delivery. Treat the result as a draft, then

either refine the prompt (if blocking is right but the realized scene is off)

or re-block with a new sketch (if the staging itself needs to change).

Two prompt patterns:

• Realization pattern — describe the scene the sketch was pointing toward,

including lighting and mood, and end with a camera direction that matches

what the sketch implies. Use this when you want a faithful interpretation

of the rough drawing.

• Variation pattern — keep the blocking implicit (the sketch handles it)

and use the prompt to swap style, lighting, genre, or time-of-day across

multiple generations. Same composition, different treatments. Useful for

picking a visual direction before committing.

Cross-reference: when the test reads well and you want a longer, multi-shot

realization, move into Cinema Studio with the validated blocking in mind. If

you need higher fidelity earlier in the process — a pre-rendered hero frame

that subsequent shots can reference — the Hero Frame workflow inside Cinema

Studio is the alternative path. See higgsfield-cinema for that handoff.

Sora 2 Trends

A templated workspace built on top of the Sora 2 model, tuned for trend-led

short-form content. The distinction worth holding onto: this is *not* the same

as selecting Sora 2 as a raw model elsewhere on the platform. Trends wraps the

model in pre-tuned templates optimized for viral pacing, vertical framing,

and quick iteration cycles. The templating is the point — it does work that

would otherwise have to be re-invented every shot.

When to use:

• TikTok, Reels, and Shorts deliverables — anywhere the format itself is more

important than full cinematic control.

• Trend-jacking, where speed of iteration and platform fluency beat per-shot

polish.

• Content where the audience expects a specific format-native rhythm rather

than a deliberate director's hand.

Distinguishing features:

• Vertical-first composition. 9:16 is the default, with 1:1 and 4:5 also

common. Compositions are framed for phones, not for film.

• Pacing optimized for short-attention spans. Fast cuts, hook-driven

openings, and a "payoff by second three" structure are baked into the

templates rather than something the prompt has to enforce.

• Templated patterns for common trend formats. POV reveals,

before-and-after flips, "tell me without telling me" setups, day-in-the-life

loops, and similar recurring structures each have a starting template.

Picking the template is most of the structural work.

Input characteristics: usually a hook idea or a trend reference. Characters

and locations can be specified, but the templating handles a lot of structure

that you'd otherwise have to direct manually. The lift on input is light.

The prompt's role: provide the specific subject and beat for the chosen

trend template. The template handles pacing and format; the prompt handles

*what* the shot is about within that format. Think of the prompt as content

slotted into a structure that's already running.

Trade-offs vs. Cinema Studio: Trends sacrifices fine-grained control in

exchange for speed and platform-native output. If the deliverable needs

deliberate camera direction, character continuity across shots, or genre

framing that the trend templates don't carry, Cinema Studio is the right

workspace and the wrong question to ask in Trends.

Cross-reference: for one-click trend-flavored workflows that overlap with

this workspace (Vibe Motion, the various style apps), see higgsfield-apps.

When full cinematic control is needed instead of templated trend pacing,

escalate to higgsfield-cinema.

Click to Ad

The product-focused ad workspace. Paste a product URL or upload an image and

Click to Ad produces a video ad variation tuned for ecommerce use. It's the

fastest way to produce packshot-adjacent content without writing a cinematic

prompt. Use it for product ads, ecommerce variations, and quick-turn commercial

content. For the full Click to Ad decision tree, companion product-ad apps

(Packshot, Giant Product, Macro Scene, Billboard Ad, and the rest of the ad

family), see higgsfield-apps.

Higgsfield Audio

The standalone voice workspace — and the most commonly misplaced one on the

platform, because Higgsfield has three distinct audio surfaces that all sound

similar by name. This workspace is the one where audio is the deliverable

itself, not a layer baked into a video and not a sync target for a generated

face.

When to use:

• Voiceover work where the final output is an audio file (narration, podcast

inserts, ad reads, e-learning tracks).

• Voice swap on an existing audio track — change who says it without changing

the timing or the emotional shape of the performance.

• Translation of an existing audio track into another language, ideally

preserving as much of the original speaker's character as the engine

supports.

• Any time you need a voice *asset* that will be fed into a downstream

workspace (a finished track to feed into Lipsync Studio, an ad read to drop

under a Click to Ad video, narration to underlay a Cinema Studio sequence).

Three core capabilities:

• Voiceover generation. Text in, spoken audio out. Voice characteristics

— gender, age range, tone, pacing, energy level — are directable. Use this

for greenfield voice work where there is no source audio.

• Voice swap. Source audio in, same audio in a different voice out. The

timing, prosody, and emotional inflection of the original performance carry

over; only the speaker identity changes. This is the right tool when an

existing performance is structurally good but the voice is wrong (a bad

scratch track, a placeholder read, an unlicensable original).

• Translation. Source audio in, the same content rendered in another

language out. Where the engine supports it, the original speaker's voice

characteristics are preserved across the translation, so the localized

version still sounds like the same person.

Input characteristics:

• For voiceover: clean text or a script. Stage directions and tone notes

belong in the prompt, not in the script body.

• For voice swap: source audio plus a target voice reference. The cleaner the

source, the cleaner the swap.

• For translation: source audio plus the target language. Optionally a voice

match preference if the engine offers it.

The prompt's role: voice descriptors and direction notes. Tone

(calm/urgent/conversational/authoritative), age range, accent, energy, and

pacing are all directable through the prompt. The synthesis itself is handled

by the audio engine — the prompt steers performance, not synthesis.

Distinction from Lipsync Studio: Lipsync Studio puts a voice (often

generated here) onto a video character. The sync direction there is

audio-to-video — the face follows the audio. Higgsfield Audio standalone

produces voice as a deliverable in its own right, with no video involvement

required. If the user wants a talking character on screen, route to Lipsync

Studio; if they want a voice file, this is the workspace.

Distinction from in-video audio: the SCELA layer of a Cinema Studio

generation (dialogue, SFX, ambient, BGM) is *baked into* the video at

generation time. That work lives in higgsfield-audio. Standalone Higgsfield

Audio is voice-as-asset — a separate deliverable, not a layer.

Cross-reference: for the in-video audio layer design (SCELA, dialogue

blocking, ambient sound design, BGM direction), see higgsfield-audio. For

AI actor voices that pair with avatars and Soul Cast characters, see

higgsfield-soul. For when the voice asset produced here gets dropped onto a

generated face, see Lipsync Studio above.

Higgsfield Collab

Collab is Higgsfield's built-in collaboration space — the team-and-community

layer that sits across the workspaces above. It is where a campaign or scene

stops being a solo job and becomes a shared one. Use it when more than one

person needs to generate, review, or talk through work together.

Shared projects. A project is a shared space where members generate

together, chat, share assets, and hop on calls. Projects have three access

levels:

• Private — invited members only.
• Public — discoverable, view-only for non-members.
• Open — anyone can join with owner approval.

Real-time communication. Chat is built into every project. Any project

with two or more members supports audio and video calls — join, leave, and

mute from inside the project, with a persistent call bar that stays visible as

you navigate so you never lose the call.

Share to Collab. Send any generation straight into the project chat with

one tap using the Share to Collab toggle. Teammates see the output along

with its prompt, model, and progress, so they can react, give feedback, or

build on it immediately.

Community feed. A recommended-and-trending feed of published work across

the platform, for discovery and inspiration.

Orgs & Team Plans. Free Orgs enable collaborative identity sharing; paid

Team Plans unlock shared credits, shared elements, admin controls, and SSO.

Users also earn karma by liking, commenting, generating, and inviting friends,

and can exchange karma for credits.

> Collab vs Shared Canvas. Collab is project-level collaboration —

> shared spaces, chat, calls, and a community feed. Shared Canvas (see

> ../higgsfield-canvas/SKILL.md § Shared Canvas) is board-level live

> co-editing of a single node graph. Reach for Collab to organize a team and

> a campaign; reach for Shared Canvas to build one pipeline together in real

> time.

This sub-skill documents only the collaboration features Higgsfield publishes;

it does not assert limits beyond those.

The Six-Step Project Logic

Every workspace follows the same high-level project flow. Memorizing this

sequence means you can navigate any new workspace without re-learning it.

• Open the right workspace — the one matching your production problem.
• Choose the right feature inside it — mode, preset, or tool that fits the

specific shot or asset you're producing.

• Select the model that fits the job — quality, speed, and cost tradeoffs

live here, inside the workspace you already chose.

• Prepare the input — hero frame, reference image, source video, audio

clip, or any starting asset the workspace expects.

• Add the prompt, script, or controls — MCSLA for scene prompts, UI

settings for Cinema Studio, motion reference for Motion Control, and so on.

• Generate, review, iterate — expect multiple passes. Change one variable

per iteration so you can tell what drove the improvement.

Relationship to MCSLA

Workspace-first is a routing layer that sits ABOVE the MCSLA prompt formula.

Once you're inside the right workspace, MCSLA (Model · Camera · Subject · Look ·

Action) still governs how the prompt itself is built. Workspaces don't replace

MCSLA — they decide which environment you'll apply MCSLA inside. A Cinema Studio

prompt, a Sora 2 Trends prompt, and a Click to Ad prompt all benefit from MCSLA

discipline, but the UI controls, defaults, and model options around each prompt

are different. Choose the workspace first; then apply MCSLA cleanly inside it.

Related skills

• higgsfield-cinema — prompt patterns for Cinema Studio (all modes)
• higgsfield-audio — audio layer design for Lipsync Studio and in-video audio
• higgsfield-soul — character references and Soul Cast avatars, often upstream of Lipsync
• higgsfield-apps — one-click workflows, including sketch-based and trend-based entry points
• higgsfield-models — model selection once a workspace is chosen
• higgsfield-prompt — MCSLA formula applied inside any workspace