pruna-run (fast entrypoint)
Use this when the user wants immediate execution from one incoming prompt — agent routing only (no bundled CLI runner).
Before any API call: generation-diversity.md (random seed ritual / SSoT + axis rotation).
Before paid calls: requesting-generation-feedback.
When NOT to use
- Multi-scene plans with approval gates → use a workflow skill (
avatar-multi-scene,music-video, …) - Editing an existing image →
p-image-edit - Recipe menus or multi-step chains → pruna-generative-pipeline
What it does
Read the user's prompt and pick the shortest chain:
| Route | When | Chain |
|---|---|---|
| image | Still only, fastest draft | p-image |
| image-hq | Still only, high-quality photoreal | p-image-ideogram |
| i2v | Motion from a still | p-image or p-image-ideogram → p-video |
| avatar | Talking head | p-image or p-image-ideogram → p-video-avatar |
Agent execution
- Confirm
PRUNA_API_KEYis set (andREPLICATE_API_TOKENonly if the chosen route needs Replicate). - Apply generation-diversity.md — ritual seed + axis rotation before the first generation.
- Route from the table above; follow the linked tool skill for HTTP payloads and async polling.
- For avatar: draft natural
voice_script+ realisticvoice_prompt; get user approval beforep-video-avatar. - Log prompts, seeds, and output URLs in a short
manifest.jsonbeside outputs for reproducibility.
Notes
- Prefer async parallel for video/avatar — poll until
succeededper pruna-api.md. - Multi-scene work: use dedicated workflow skills — narrated films use scene-anchor-triple.md (narrated-multi-scene).
