video-prompt-craft
Skill8 starsChanged 30 days agoMITView on GitHub
Write text-to-video and image-to-video prompts that hold up in real generation. Shot-brief grammar, JSON prompting, negative prompts, and failure-mode hardening. Use when drafting, reviewing, or fixing a video generation prompt for any model.
In this repository8 other files it gives its agents.
- assembly-and-deliverySkill
- character-consistencySkill
- cinematic-directionSkill
- dialogue-and-audioSkill
- model-adaptersSkill
- muse-google-veo-webSkill
- story-and-shotsSkill
- video-qaSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Brief it: Write the prompt in the shot-brief grammar (references/prompt-grammar.md).
- Very specific: line 25
Harden it: Add the negative prompt (references/negative-prompts.md) and walk the failure-mode list (references/failure-modes.md).
- Very specific: line 28
Adapt it: Run the finished prompt through the target model's adapter (../model-adapters/SKILL.md).
- Distinctive: line 38
One variable at a time: When a result is wrong, change one slot and regenerate.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video Prompt Craft
Most bad AI video comes from bad prompts, not bad models. The usual failure is vagueness: "a beautiful cinematic shot of a woman walking in Paris." That sentence asks the model to invent the shot, the walk, the light, and the city. It will invent all of them differently every time.
This skill teaches one habit: say what the camera sees, beat by beat, and say what it must not do. Everything else is commentary.
The workflow
- Brief it. Write the prompt in the shot-brief grammar
(
references/prompt-grammar.md). Eight slots, no more. If a slot is empty, that's a decision you're handing to the model. Do it on purpose or fill it. For image-to-video, subtract instead:references/i2v-prompting.md. - Consider JSON. For pipelines, revisions, or multi-shot work, write the
brief as JSON instead of prose (
references/json-prompting.md). Same information, but diffable and harder to silently change between drafts. - Harden it. Add the negative prompt (
references/negative-prompts.md) and walk the failure-mode list (references/failure-modes.md). Restrictions are cheap; reshoots are not. - Adapt it. Run the finished prompt through the target model's adapter
(
../model-adapters/SKILL.md). The grammar is universal; every model has quirks.
Rules that don't bend
- Concrete verbs beat adjectives. "She strides" beats "she walks beautifully." "Rain hammers the awning" beats "a moody rainy atmosphere."
- Counts and positions are load-bearing. "Three seagulls, left to right" beats "some birds." Models drop or multiply anything you don't pin down.
- One variable at a time. When a result is wrong, change one slot and regenerate. Changing three things teaches you nothing.
- Reuse character descriptions verbatim. Paraphrasing a face between shots
is how you get two different people. Copy-paste, don't rewrite. (Full
discipline:
../character-consistency/SKILL.md.) - Time the beats. A 10-second clip needs its action budgeted in seconds, not vibes. If you can't split the action into timed chunks, the prompt isn't ready.
When this skill is not enough
- Camera language (angles, moves, lenses):
../cinematic-direction/SKILL.md - Multi-shot structure, storyboards, review gates:
../story-and-shots/SKILL.md - Dialogue, voiceover, subtitles, mixing:
../dialogue-and-audio/SKILL.md - Stitching clips and exporting:
../assembly-and-delivery/SKILL.md - Checking the result:
../video-qa/SKILL.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

