asset-generation

# Image / Video Asset Generation ## Purpose Generate standalone image or video assets from Creatify's model catalog — b-roll, product shots, cinematic clips, edited/upscaled images, etc. For a talking avatar use `avatar-video` (Aurora) instead; this skill is for general generation models. ## Mandatory discovery flow — never guess params 1. **`asset_list_models`** — browse the catalog (~90 models). Filter with `media_type` (`image` | `video`) and/or a `query` substring. Each row has `model_name`, `media_type`, `title`, `description`, `plan_level`. 2. **`asset_get_model_schema(model_name)`** — get that model's `input_params`. **Do NOT guess input_params** — they differ per model. 3. **`asset_generate`** — start the generation with the model + validated params. 4. **`asset_get`** — poll until the asset is ready, then return its URL. ## Picking the model - Match `media_type` and the **task variant encoded in the model name suffix** to what the user has: - `/text-to-video`, `/text-to-image` — prompt only. - `/image-to-video`, `/image-to-image` — a starting image. - `/reference-to-video` — one or more reference images for subject/style. - `/first-last-frame-to-video` — a start and end frame. - `/edit`, `/video-to-video/edit` — modify an existing image/video. - `/extend-video` — lengthen an existing clip. - Respect `plan_level` (`free` / `starter` / `pro`): if the user's plan can't use a model it will be gated — surface that instead of failing silently. - The catalog changes over time. Always discover at runtime with `asset_list_models`; do not rely on a hardcoded model list. ## Align Before Generating Generation is paid and time-consuming. Confirm the load-bearing choices before generating: what to generate, image vs video, aspect ratio/duration, and (if it matters) which model tier. Establish one sample before batching many. ## Async & Verify `asset_generate` is long-running. Poll `asset_get`; only report success once it returns a finished asset URL. See `verification` and `known-errors`.