CinePrompt
“A nice shot of a woman in a room” gets you exactly that — flat, eye level, evenly lit. Not because the model is weak, but because nothing asked for anything else. Every tag here says what it does, so choosing one is also learning what it is called.
Subject
Say what is happening. The tags below say how it is shot — keep the two separate and both get clearer.
Shot size
How much of the subject is in frame.
Camera angle
Where the camera sits relative to the subject.
Lens
Focal length. Changes the relationship between subject and background.
Aperture
How much is in focus.
Movement
What the camera does during the shot.
Light quality
Hard or soft. This is what makes an image feel expensive.
Light direction
Where the key comes from.
Time of day
Sets colour temperature and shadow length.
Palette
The colour relationship of the whole frame.
Format
What it was notionally shot on.
Composition
How the frame is arranged.
Mood
The feeling. Vaguest, so it goes last.
Compiled
Describe the subject, then pick how it is shot.
Ordered by department rather than by the order you clicked — a model weights early tokens more heavily, so the subject leads and mood trails. The same selection always compiles the same way.
On the dialects
This compiles a plain prompt you can paste anywhere. Inside the Studio you do not need it — a shot carries structured data, and the Router rewrites it into whichever dialect the chosen model speaks:
seedance — Acoustic markup and camera velocity. Generates audio in the same pass as picture.
kling — Labelled blocks. Suits its multi-shot structure.
veo — Flowing prose, no markup — brackets would render as visible text.
tagged — Dense tags for open-weight models. Dialogue is dropped: they have no audio and would draw the words.
prose — The safe default every model tolerates.