How to Write an AI Video Prompt That Does Not Fall Apart
11 Nov 2026 · 1 min read · by the SkillMomo team
Most AI video fails come from the same problem: the prompt describes a mood, not a shot. "A beautiful cat in a cosy room" gives the model nothing to film.
The five parts we use
- Format - aspect ratio and style. "Vertical 9:16, realistic" or "16:9 documentary footage".
- Subject - who or what, with two or three physical details. Colour, clothes, size.
- Action - one clear action, then one reaction. Models struggle with more than two beats in 8 seconds.
- Camera and light - lens, movement, time of day. "50mm, slow push-in, warm window light."
- Sound and length - for tools with audio, name the sounds. Always end with a duration.
Before and after
Before: A funny dog in a kitchen.
After: Vertical 9:16, realistic. A beagle stands on a kitchen chair, sniffs a plate of rotis, looks left and right, then quietly takes one and freezes when the owner walks in. Soft morning light, 35mm, handheld. 6 seconds.
The second prompt has a beginning, a middle and an end. That is what makes people watch twice.
Small habits that help
- Keep one prompt to one location.
- If a detail keeps disappearing, repeat it once near the end.
- Generate three versions before you change the prompt. Randomness is real.