tribgames/mixdog/video

Video generation (media tool · Media Studio)

Use media to generate short video clips; read this guide before the first call. The media tool schema and list own the controls, lane catalog and sign-in state; nothing here names a lane or a model. A still the clip needs first is the image skill's job.

1. Which route

Hard rule — video only when motion was asked for: a deliverable never gets a clip by default; a still that carries the meaning is not turned into a video.

Hard rule — no signed-in lane, no clip: when list (or a failed generate) reports no video lane, say so and stop; never substitute a slideshow, a GIF from stock, or a placeholder. Default — one generation per job, one corrected retry: a miss is retried once with a single changed variable and every invariant restated; a second miss is reported, not attempted a third time. Several candidates only when the user asks.

2. Call order

1. media action:'list' kind:'video' once per session: signed-in lanes, models, and remembered — the lane/model a generate runs on when none is passed. Add model: for that model's controls; pass a control only when that model lists it.
2. Hard rule — pick the source in this order: a source the user names is passed as lane (and model if named), resolved from list; if it is not signed in, say so rather than swap. Otherwise omit lane/model; the runtime uses the configured default (pinned), else the current Studio selection or last clip's lane/model (remembered), else the signed-in ChatGPT lane if it makes video, else the first signed-in lane, so the clip comes from what the user chose. Choose a lane/model by capability only when the job needs one the selected model lacks (reference count, a duration outside its range).
3. Before writing the prompt, read references/prompting.md.
4. media action:'generate' kind:'video', prompt, path, aspect, duration, resolution, references, wait:true. Use wait:false only when background execution is requested or useful independent work remains. The result reports the lane, model, and laneSource (requested = a lane explicitly selected in the call (a model alone is not), pinned = the user's configured default for video, remembered = their last choice, first = the default lane or the first available signed-in lane) that actually ran.
5. A background or resumed job: media action:'status' job:<id> path:<file> at spaced intervals; the call that sees done writes the file. A running job is pending, not completion and not a failure: do not regenerate it or tight-loop status calls; cancel only when the user moves on.
6. Hard rule — look before using: frame-sample or play the file; check the opening frame, that the one change happens, the ending frame, and that no text, logo, or extra subject appeared.
7. The result is also a Studio asset (assetId); the file at path is the copy the work uses.

3. Prompt and controls

The prompt is an active shot, not a list of elements: labeled items, never keywords or JSON. A detailed user prompt is normalized and nothing is added; a generic one gets only shot design, camera intent, and a concrete environment — never extra characters, props, brands, or story beats. Exclusions: no text, no logos, no watermarks; no faces only for a field for copy or on request.

• aspect from the destination (a feed or story clip 9:16, an embed or presentation 16:9, a tile 1:1), a listed value; the still generated first uses the same aspect.
• duration fits the beats and the model's list or range. References beyond one shorten the allowed duration on most models — read the model's controls before promising a length.
• resolution: lowest listed for a draft that checks motion and pacing, highest for delivery.
• references: take file paths (cap: maxReferences): the user's own file, a still from the image skill, or an earlier clip's frame — never a downloaded third-party image. Each gets an index and a role in the prompt; identity references (face, product, logo) go first.

4. Saving, reporting, provenance

• Never overwrite: an existing file gets a sibling (clip-v2.mp4) unless replacement was asked; output goes beside the work that uses it or where the user named.
• Report the final path, the lane/model that actually ran, duration, resolution, aspect, and the prompt.
• Declare generated media where it lands (caption, note, or message) with lane/model and prompt; never present it as footage of a real event, place, or person.

5. Failure and cost

A generate error carries lanes (signed in for video) or available (models): choose from it and call again, or stop. Video costs far more credit than a still: one call per job, drafts at low resolution, no speculative variants.

Resources

references/prompting.md — read before constructing a video prompt, and when extending a clip.