Vertical Short Maker · публичный навык Open Skill

Vertical Short Maker

Vertical Short Maker — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.

@madriz96

Anime

15 4 13

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4168388/2026-09-27/vu4b82xb.jpg

# Vertical Short Maker

## When to use
- The user gives a single phrase / idea and wants a vertical (9:16) short video.
- The user uploads an image, video, or audio file and wants a vertical short made from it.
- The user wants to continue, restyle, or refine a previously made short.

## Steps
1. **Recall memory first.** Call `memory_query` (and `memory_whoami` if available) to recall the user's saved preferences (style, tone, duration, platform, voiceover). If nothing relevant is found, don't invent — ask briefly or use sensible defaults (portrait 9:16, ~5–10s).
2. **Determine the input type** and pick the generation path:
   - **Text phrase only** → `generate_video` with `model='t2v'`, `orientation='portrait'`, `prompt` = the phrase expanded into a vivid scene description.
   - **Uploaded image** → `generate_video` with `model='i2v'`, `input_urls=[image]`, `orientation='portrait'`; describe the desired motion in `prompt`.
   - **Uploaded video** → first `recognize_video(video_url=<asset>, prompt="Describe the plot, key shots, and style")` to understand it, then generate a new vertical short via `t2v` (or `i2v` with a key frame) matching that content.
   - **Uploaded audio** → first `transcribe_audio(audio_url=<asset>)` to get the speech/lyrics, then generate a vertical short via `t2v` from that content.
3. **Generate the video.** Always `orientation='portrait'` (9:16). Pass `name` = a short title for the clip and `entity_type='final_videos'`. Apply the user's remembered style/tone/duration preferences to the `prompt`.
4. **Save preferences.** After the result, `memory_store` any new durable preferences the user expressed (style, tone, preferred duration, platform, voiceover) with `namespace='vertical-short-maker'` and an appropriate `kind` (`preference` / `project` / `person`).
5. **Reply** in natural language summarizing the short; don't paste large URLs. Offer next steps via `set_guidance` (e.g. restyle, make another, adjust duration).

For multi-step runs, first use `write_todos` to list the steps and update it as you progress.

## Handling "forget/correct"
- To forget a saved preference, `memory_query` to find its `id`, then `memory_forget(id)`.

## Notes
- Video generation does not support NSFW/adult content on any channel — politely decline such requests.
- If generation fails or returns an error, report it honestly and offer to retry with adjusted parameters; never fabricate a video URL.
- Intermediate step-by-step products (if any) will be visible on the canvas.
- Memory is user-level: other skills may also read these memories.