Why most AI video prompts look flat
The default AI video clip is bright, evenly lit, and centered, three things real cinema almost never is. The fix is rarely a bigger model. It is a more specific prompt that names the camera, the lens, the light, and the movement.
1. Name the lens and the focal length
Generic prompts produce generic depth. Tell the model you want a 35mm lens at f/1.4 and it starts separating subject from background the way a real camera does.
- Wide lenses (24-35mm) exaggerate space and motion.
- Long lenses (85mm and up) compress distance and isolate subjects.
- Fast apertures (f/1.4 to f/2.0) give you soft, cinematic bokeh.
2. Describe the light, not the mood
Moody means nothing to a model. A single practical lamp camera-left, deep shadow on the right half of the face means something. Light direction is the single biggest lever you have.
3. Move the camera on purpose
Static frames read as stock. A slow dolly-in builds tension; a locked-off tripod frame reads as observational. Pick the movement and name it.
Five prompts you can steal
Each of these is written to push the model away from the flat, over-lit default and toward something that reads as film.
- 35mm, f/1.4, slow dolly-in on a subject lit by a single window, film grain, muted teal shadows.
- Handheld 50mm, overcast daylight, subject walking toward camera through shallow fog.
- Locked-off 85mm portrait, single warm practical lamp camera-left, deep negative fill right.
- 24mm wide, low-angle, golden-hour rim light, dust in the air, anamorphic flare.
- Macro 100mm, racking focus from foreground object to background subject, hard single source.
Make it consistent
The prompt gets you one good clip. To turn clips into a scene, keep the lens, the light direction, and the grain identical across every shot in the sequence. Avocadio handles that consistency for you, the same look across every generated clip in a project.
Related guides