How to Describe Light to an AI Video Model
Video models read light the way a gaffer talks about it: where the source is, how hard it is, what colour it is, and how deep the shadows go. "Cinematic lighting" tells a model nothing. "Hard warm key from frame right, high; soft cool fill from frame left; low-key" tells it exactly what to draw.
Direction
Say where the key comes from relative to the camera: from the front, three-quarter, side on or from behind. Side and three-quarter light give a face shape; light from behind gives a rim that separates the subject from the background.
Height
High light reads as day or a ceiling source; eye-level light feels intimate or staged; light from below reads as unease or firelight.
Quality
Hard light (sun, a bare bulb) throws crisp shadows. Soft light (overcast sky, a big bounce) wraps around the face. Most moods are set by this one choice.
Colour and contrast
Name the colour as a source would have it: warm tungsten, cool daylight, moonlight blue, magenta neon. Then the ratio: low-key means deep shadows and little fill; high-key means open, bright shadows.
One light logic per shot
Pick one coherent setup per shot and keep it across the shots of a scene. Mixed, unmotivated sources confuse the model and make cuts jump.








