Give an edited video and its final SRT to Codex, Kimi, or Claude Code, then say a few words about the result you want. MotionTalk reads the content, proposes a director plan, and completes production after one approval.
You do not need to understand video engineering or fill in complex parameters. Give the assets to Codex / Kimi / Claude Code and describe the result as if you were talking to an editor: landscape or portrait, presenter placement, PPT or screen recording, caption feel, and any animation you want.
Give the Agent your video and subtitles, then describe the result → The Agent proposes a director plan → Receive the final video → Optional adjustments
For example:
Use MotionTalk for this video and its subtitles. Switch or interweave the
presenter, full-screen MG, and other video B-roll wherever you think it best
serves the content.
After approval, the Agent completes production, checks the result, and delivers the final video. Stop when it feels right, or describe any adjustments in plain language.
The four effects below are starting points in one standalone reference prompt. Naming one can quickly produce a similar video, but they are not a required four-way choice or the boundary of MotionTalk. Describe the desired layout, captions, presenter, screen recording, or animation directly in natural language whenever you want a different result. Every frame is from a real local delivery.
 1. floating-overlay Keep the presenter or recording full-screen and add only lightweight MG in safe zones. |
 2. mg-with-presenter-window Make MG, screenshots, or recordings primary while the presenter remains in a circle or rectangle. |
 3. switching Switch between presenter, full-screen MG, and recordings so each state stays readable. |
 4. PPT Focus Portrait Use PPT as the portrait primary visual, with self-fitting captions below, a bottom-right presenter, and progress. |
All four guides live in
references/04-reference-theme-prompt.md.
The prompt only helps the Agent understand a direction faster; the final plan
still follows the current assets and natural-language request.
- Understand before designing: the Agent reads the full video and subtitles instead of adding random effects;
- One approval, continuous delivery: after the director plan is approved, production continues without repeated interruptions;
- Designed for each video: visuals follow the content and your request, not a fixed template;
- Checked before delivery: captions, presenter proportions, occlusion, source matching, and the final video are reviewed;
- Reference-theme prompt: align quickly with four real examples, or skip themes entirely and design the current project in natural language.
npx skills add PoetCoderJun/MotionTalkpython3 -m unittest discover -s tests -v
node scripts/render_master.mjs --help
node scripts/validate_master.mjs --helpOriginal repository material is available under CC BY-NC-SA 4.0 for non-commercial use. Commercial use requires separate written permission.