cinematic-direction
Skill8 starsChanged 29 days agoMITView on GitHub
Direct the look of AI video like a cinematographer: camera language, movement, lighting, color, and style lenses. Use when a prompt needs a real shot in it, not just a subject, or when results look flat and amateur.
In this repository8 other files it gives its agents.
- assembly-and-deliverySkill
- character-consistencySkill
- dialogue-and-audioSkill
- model-adaptersSkill
- muse-google-veo-webSkill
- story-and-shotsSkill
- video-prompt-craftSkill
- video-qaSkill
The file
Files
Cinematic Direction
AI video models know what things look like; they don't know what a shot is unless you tell them. "A man in a cafe" gives you a man in a cafe. "Low-angle medium shot, slow push in, man in a cafe, sodium streetlight through blinds" gives you cinema. The difference is entirely in the directing vocabulary.
This skill is that vocabulary, organized the way a DP thinks: camera first, then light, then the overall style.
The workflow
- Pick the shot. Framing + angle + movement from
references/camera-language.md. One of each, stated plainly. If the camera doesn't move, write "locked off". Stillness is a decision. - Light it. Time of day, source, quality from
references/lighting-and-color.md. Light does more for "cinematic" than any other single choice. - Set the style. A style lens from
references/style-lenses.md: a film stock, an era, a director's grammar, stated as concrete tokens, not adjectives. One lens per project, used consistently. - Write it into the brief. Camera and light slots in the shot-brief
grammar (
../video-prompt-craft/references/prompt-grammar.md) are where this lives. This skill supplies the words; prompt-craft supplies the sentence.
Rules that don't bend
- Name the lens, don't describe it. "35mm, f/2, Kodak Vision3 500T" beats "shot on film, cinematic look." Signature tokens (stocks, lenses, eras) are how models actually key into looks.
- One move per shot. Push in or orbit or track. Two moves in 8 seconds reads as indecision and usually smears.
- Motivate the move. The camera moves because the story does: push in on a realization, pull back on isolation, track with pursuit. Unmotivated movement is the fastest way to look like a screensaver.
- Light is story. Hard noon sun is not the same scene as blue-hour window light. Choose light for what the scene means, then let the model handle the photons.
- Restraint reads as quality. Fewer style tokens, used consistently, beat a soup of "cinematic ultra-detailed 8k masterpiece." Every adjective you cut makes the remaining ones stronger.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

