mcpbeat Sign in

Flux 3 Cinematic Inserts Agent Skill

Use when writing a FLUX 3 text-to-video prompt for a standalone shot. Covers shot craft, reliable concept patterns, camera, continuity, and review.

3k tokens
context cost
the whole folder, loaded on every use
5
files
instructions only
0
copies elsewhere
how many repositories repackaged it
106
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/black-forest-labs/skills --skill flux-3-cinematic-inserts

What comes with it

8 964 bytes besides the instruction
metadata.json
references/field-guide.md
references/review-scorecard.md
templates/brief.json

The instruction itself

1 sections, as written by the author

FLUX 3 Cinematic Inserts

Build one standalone shot around one motivated physical or atmospheric event: B-roll,

a bridge, a material detail, a capability test. Reroute when success requires supplied

media (flux-3-keyframes-continuation), exact typography, frame-accurate sync, a

multi-step mechanism, close human performance, or finished ad copy; those need

conditioning or post, not a text-only generation.

Before writing, settle the contract in one sentence: the visible cause, the physical

response, and the payoff. Then pick a proven lane from

references/field-guide.md (atmospheric light in a medium,

wet reflections, viscous macro, tactile contact and ignition, restrained noir); treat

anything it marks experimental or parked as a test, not a promise.

Write the shot in this order:

  • Opening frame: composition, subject, setting, light.
  • Visible cause: source and contact point in the same frame as the response.
  • Physical response: direction, material behavior, restraint.
  • Payoff: one final state that works as an edit point or a still.
  • Continuity contract: same set and geometry; no cuts, transitions, frozen or

repeated frames, or resets.

  • Audio intent: name material textures; route critical sync and mixing to post.

Camera is one physical contract: framing, angle, movement, focus. A locked-off camera

cannot track; an extreme close-up cannot establish. Locked and deliberate by default;

move only when it serves the event.

Style must be visible: "tense" becomes rigid posture, shallow breathing, and a slow

push-in; "vintage" becomes a named capture format, grain structure, and practical

light. No unmotivated sparks, bells, or effects.

Draft first (draft: true) while the concept is unproven, and judge the draft on event

legibility, composition, camera, and continuity; softness is the low-step trade and

resolves at full quality. Iterate as a ladder, one consequential change per variant:

event geometry, then material or light, then composition, then duration. After two

structurally similar misses, change the concept or the production method, not the

adjectives. Review the result with

references/review-scorecard.md: a technically valid

clip still has to be useful, physically legible, deliberate, and interesting.

Hand the finished prompt to flux-3-generate as a text-only request, with the risks and

anything post must own.

How to use it

Copy the folder

Take black-forest-labs/flux-3-cinematic-inserts from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.