AI Video Prompt Guides: How to Write Prompts That Work in Each Tool
Tool-by-tool prompt guides, tips and tutorials for AI video, image and audio generators. Every guide is built on generations we ran ourselves, including the ones that failed, and links to the vendor’s own documentation.
The short answer
Most AI video tools reward the same habits: a fixed structure, a time budget for every shot, motion written as concrete events, and reference images that each do one job. What differs is the syntax and the failure modes, which is what each guide below covers.
Guides by tool
Seedance 2.5
Multi-shot prompts with labeled sections, seconds per shot, real-time motion, wheel and fabric physics, and how to use reference images without copying their flaws. Includes our test results from October 1, 8 and 9, 2026.
The Higgsfield link is an affiliate link: we earn a commission if you subscribe through it, at no extra cost to you. Higgsfield resells several models we score separately, Seedance among them.
Six habits that work in every video generator
- Structure before style. Say how many shots, how long each is, and where the cuts fall, before any adjectives about the look.
- Events, not moods. “She brakes and puts her foot down at 9.5 s” gives the model something to render; “a calm, cinematic moment” gives it time to fill.
- Numbers where they exist. Speed, height, lens, seconds. Models follow figures more reliably than comparatives.
- Physics in writing. Wheels, cloth, liquids and carried objects need a sentence about how they move, or the model falls back on defaults.
- One job per reference. Tell the model what each image defines and what it must not copy from it.
- Draft cheap, then finish. Test the wording at the lowest resolution, then commit the final length and quality.
How we build these guides
Each guide combines the vendor’s own documentation with generations we ran ourselves, and every first-hand result is dated. When a tool changes, we re-test before we update the guide. Sample prompts are written by us for the guide; we link to the guides we learned from rather than copying their examples.
Found a prompt that behaves differently? Tell us in the forum or in the comments under the guide.