Fast AI Video Maker · публичный навык Open Skill

Fast AI Video Maker

Fast AI Video Maker — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.

@rmworldof

ShortDrama

3 0 4

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3762060/2026-09-25/6xsbtrcm.png

# Fast AI Video Maker

## When to use
- The user wants a short AI video made quickly from a one-line idea, a short script, or an uploaded image/photo.
- The user wants a fast vertical short (TikTok/Reels/Shorts style) without a complex multi-stage cinematic pipeline.
- The user wants a quick clip with optional background music, stitched together.

This is the "fast and simple" video maker. For complex cinematic drama, trailers, or music videos, the user should use a dedicated cinematic/music-video skill instead.

## Steps
1. **Clarify the input** (briefly, don't over-ask): a one-line idea, a short script, or an uploaded image. If the user gives only a vague idea, ask one short question to pin down the subject and mood. Default to a **vertical (portrait 9:16) ~10s short** unless they say otherwise.
2. **Break the idea into 2–3 short shots** (each ~3–5s). For each shot, write a vivid visual description (subject, action, setting, lighting, mood). Keep the subject consistent across shots.
3. **Generate a keyframe image per shot** with `generate_image`:
   - `model='t2i'` (text-to-image) when there's no reference image; `model='i2i'` when the user uploaded a reference image/photo to keep the subject.
   - `orientation='portrait'` (9:16 vertical), `resolution='720P'`.
   - `name` = shot title (e.g. "shot1_open"), `entity_type='storyboard_images'`.
4. **Animate each keyframe** with `generate_video`:
   - `model='i2v'`, `input_urls=[keyframe_asset_id]`, `orientation='portrait'`, `resolution='720P'`, `duration=5`.
   - Prompt = the motion/animation instruction for that shot (keep subject identity consistent).
   - `name` = shot title (e.g. "clip1_open"), `entity_type='storyboard_videos'`.
5. **Optionally generate background music** with `generate_music` (`model='music'`) matching the mood (e.g. "upbeat electronic", "calm ambient"). Keep it short (~10s) to match the video. If the user didn't ask for music, skip this step.
6. **Stitch the clips** with `mix_clips`:
   - `clips` = one entry per shot, each with `video.url` = the clip asset_id (concatenated in order). If music was generated, add `audio.url` = the music asset_id to the first clip (or as a separate track).
   - `orientation='portrait'`, `name` = final title, `entity_type='final_videos'`.
7. **Deliver**: show the final video and give a one-line summary of what was made. Offer quick follow-ups (change a shot, swap music, make it longer).

## Notes
- **Speed first**: keep to 2–3 shots and ~10s total by default. Don't add extra stages (voiceover, subtitles, complex refs) unless the user explicitly asks.
- **Consistency**: reuse the same subject description and, when possible, the same keyframe image across shots so the subject stays recognizable. For a single-subject short, prefer generating one keyframe and animating it, or use `i2i` on the first keyframe to derive the next.
- **Channels**: use bare `t2i`/`i2i` for images and `i2v` for image-to-video — never hard-code physical model names. Video generation does not support NSFW/adult content on any channel; decline such requests.
- **Honesty**: if a generation fails or a clip looks off, say so plainly and offer to regenerate that shot rather than pretending success.
- **Intermediate assets**: the keyframe images and individual clips will appear on the canvas alongside the final video — that's expected, not an error.