svenkatreddy/universal-video-skills / skills/video-prompt-craft
video-prompt-craft
VerifiedWrite 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.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.8 KB |
| references/failure-modes.md | markdown | 2.9 KB |
| references/i2v-prompting.md | markdown | 3.5 KB |
| references/json-prompting.md | markdown | 2.8 KB |
| references/negative-prompts.md | markdown | 2.0 KB |
| references/prompt-grammar.md | markdown | 3.4 KB |
references/prompt-grammar.md
View on GitHub# The Shot-Brief Grammar
One shot, one brief. Eight slots. Write them in this order; it matches how a
director actually thinks through a shot.
```
[DURATION + ASPECT]. [SUBJECT] performs [VISIBLE ACTION] in [SETTING].
Camera: [FRAMING + MOVEMENT].
Beats: [TIMED ACTION BREAKDOWN].
Audio: [DIALOGUE + FOLEY + AMBIENCE].
Preserve: [IDENTITY / GEOMETRY / LAYOUT TO HOLD].
Avoid: [FAILURE MODES].
```
## What goes in each slot
**Duration + aspect.** `8s, 16:9.` Models have hard caps (see the adapters).
State both so nobody assumes. If your story needs 30 seconds, that's four
briefs, not one.
**Subject.** Who or what the shot is about, described once, physically. Age,
build, clothing, distinguishing marks. This exact text gets reused verbatim in
every shot the subject appears in.
**Visible action.** Only what the camera can see. "He looks nervous" is not
visible. "He taps his thumb against the mug, twice, without drinking" is.
Inner states have to be translated into bodies and objects.
**Setting.** Place, time of day, weather, key props. Enough that a location
scout could find it. "A laundromat at 2 a.m., fluorescents buzzing, one dryer
still tumbling" beats "a laundromat."
**Camera.** Framing plus movement, in that order. "Medium close-up, slow push
in." If the camera doesn't move, say "locked off". Static is a choice, and
models drift when you don't choose. (Full vocabulary:
`../../cinematic-direction/references/camera-language.md`.)
**Beats.** The duration split into timed chunks: `0–3s: she enters frame left;
3–6s: she stops at the counter; 6–8s: she looks up, directly into lens.`
Keep chunks under ~5 seconds. Anything longer in one chunk will smear.
**Audio.** Dialogue as quoted lines with the speaker named, then foley, then
ambience. `MARA (quiet): "You came." Foley: bell dings. Ambience: rain on
glass, distant traffic.` If the model doesn't do audio, this slot still helps:
it forces you to decide what the scene *sounds* like, which sharpens what it
looks like.
**Preserve.** What must not change within or across shots: face identity,
product geometry, the count of people, screen direction. This is where you
fight drift before it happens.
**Avoid.** The failure modes you're pre-empting: `no morphing faces, no extra
fingers, no text on signs, no camera shake.` See `negative-prompts.md` and
`failure-modes.md`.
## A worked example
> 8s, 16:9. MARA (34, sharp jaw, copper bob, olive bomber jacket) crosses a
> night market in the rain. Camera: tracking shot at waist height, keeping
> pace on her left. Beats: 0–3s: she weaves between stalls, neon signs sliding
> past; 3–6s: she slows at a dumpling cart, steam washing over her face; 6–8s:
> she looks up into the lens and smiles, barely. Audio: sizzle and chatter,
> rain on canvas awnings, a vendor shouting in Cantonese. Preserve: Mara's
> face and jacket identical throughout; three stalls on her right, no more.
> Avoid: morphing faces, extra limbs, readable text on signs, slow motion.
Notice what's missing: "cinematic," "beautiful," "8k," "masterpiece." Those
words do nothing. The brief above is longer than a one-liner and shorter than
a paragraph of adjectives, and it gives the model decisions instead of
homework.
## Iteration discipline
Change one slot per regeneration. Keep every version. When something finally
works, the version history *is* the documentation. It shows which slot fixed
which problem.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |