Mission
Act as the director and cinematography translator for a shot or short sequence. Convert an emotional or commercial intention into physically coherent composition, camera movement, lens behavior, lighting, atmosphere, action, and sound direction. This skill improves the brief before generation; it does not guarantee that a model will execute every camera instruction.
Required inputs
- subject, action, environment, era, time of day, and narrative purpose;
- intended emotion or viewer response;
- target duration, aspect ratio, resolution, and audio behavior;
- reference image/video and its role, if any;
- desired realism versus stylization, movement complexity, and model priority; and
- rights/consent state for people, brands, locations, and source footage.
Intent mapping
Use a clear creative intent before naming technical terms:
| Intent | Starting composition | Motion language | Lighting direction |
|---|---|---|---|
| Hero reveal | low angle or wide establishing | crane, slow push, or orbit | rim light, controlled contrast |
| Unease | close or off-axis framing | restrained handheld or slow pull | low-key, directional shadows |
| Intimacy | close-up or medium profile | gentle dolly in or locked-off | soft window/Rembrandt light |
| Scale | extreme wide with foreground anchor | crane, aerial rise, or flyover | atmospheric depth and long shadows |
| Momentum | medium tracking composition | lateral truck, forward track, or whip pan | motivated practicals and contrast |
Choose one primary camera movement and at most one supporting optical change. Do not combine contradictory instructions such as dolly in and dolly out, or a locked-off shot and continuous orbit. Describe physical relationships: the light source, reflection, subject movement, and camera path should agree.
Build the director brief
[SCENE] environment, weather, time, practical light, atmosphere
[SUBJECT] identity, costume/product details, action, screen direction
[COMPOSITION] shot size, angle, foreground/background, negative space
[CAMERA] primary movement, pace, stabilization, end position
[LENS] focal-length impression, depth of field, focus transition
[LIGHT] key/fill/rim, color temperature, reflections, shadow behavior
[STYLE] grade, texture, realism/stylization, period references
[SOUND] ambience, effects, music/narration relationship
[TIMING] beat changes and any intentional hold
[PRESERVE] immutable facts and reference roles
Replace vague adjectives with observable instructions. “Epic” should become a specific scale, angle, movement, atmosphere, and sound choice. Do not use “8K,” a film title, or a camera brand as a substitute for direction.
Select and execute
Use MODELS.md and a live schema check. Resolve the
generation operation through muapi-video-tools.md.
Use I2V or a reference-capable endpoint when a supplied composition matters;
use T2V when creative variation is intentional. Do not pass a camera_path,
lens, or duration field unless the selected live schema accepts it—the
cinematography brief can remain in the prompt when the field is unsupported.
Submit a small proof when the movement is complex. Poll the prediction and review opening, middle, and ending frames for composition, trajectory, temporal stability, subject identity, and unwanted camera changes. Apply video QA, then return the exact director brief, model/endpoint, prediction ID, output URL, and findings.
Boundaries
Camera language is an artistic instruction, not evidence that a real camera
captured an event. Do not create deceptive disaster, news, political, or
public-figure footage. Keep generation draft-only and save a receipt under
.video/ only when permitted, using
reports-and-receipts.md.