20 Open-Source Repos That Let AI Edit an Entire Video for You: You Only Write the Brief
Fable 5.2 can write code to generate animated videos, but what you get from a one-line prompt is just the AI's taste. seeco compiled a list of 20 open-source repos, covering everything from editing to design to rendering; combined, they form a zero-cost AI video production pipeline. This article breaks down the role of each tool and why the most valuable part is actually writing the brief.
Fable 5.2 was released a week ago, and feeds were full of AI-generated films, games, and artwork. But few people posted a genuinely reusable workflow: which tools to use, how to combine them, and how much it costs.
seeco compiled a list on X: 20 open-source repos covering every step from editing and design to rendering. Paired with Claude, they can handle all the “clicking” operations; you only need to enter a brief.
First, some background. Fable 5.2 is an Anthropic model focused on 3D, hardware, and motion effects. It doesn't directly generate an mp4; instead, it writes a self-running HTML page. You open it in Chrome, and the animation plays. To export a video, use a script to take frame-by-frame screenshots and feed them to ffmpeg for compositing. Throughout the process, AI is both director and editor.
The value of this list is that it connects professional software, automation scripts, and design systems all to AI. Each repo solves one specific step.
Editing and Post-Production
- DaVinci Resolve MCP: Lets Claude directly control DaVinci Resolve, essentially giving AI a pair of hands to operate professional editing software.
- auto-editor: Automatically cuts out silent sections, providing a rough cut first.
- PySceneDetect: Automatically detects scene changes and works with auto-editor to do intelligent rough cuts.
- MoviePy: A Python video editing library, suitable for scripted batch processing.
- Remotion: Write videos in React, a programmer-friendly way to create motion effects.
- video-use: Upload raw footage, and it automatically removes padding, color grades, burns in subtitles, and outputs the final mp4.
Design and UI
- Magic MCP: Generates UI directly in Claude Code, similar to v0.
- Figma Context MCP: Feeds Figma design files directly to the agent.
- Penpot: An open-source Figma alternative with built-in MCP.
- agent-browser: Lets the agent open pages itself and check the design results.
- ui-ux-pro-max: 80 UI styles, color palettes, and font pairings.
- /canvas-design: An official Anthropic skill for posters and artwork.
- /pptx: An official Anthropic skill for generating PPTs.
- /theme-factory: 10 themes for slides, documents, and landing pages.
- stitch-skills: A Google Labs skill that turns prompts into UI screens.
- /frontend-design: An official Anthropic skill that generates UI that doesn't look AI-made.
- DESIGN.md: Google's format for describing visual identity to agents.
- Astryx: Meta's open-source design system built for agents.
- Excalidraw: Hand-drawn-style diagrams and wireframes.
- tldraw: An infinite-canvas SDK.
No matter how powerful the tools on the list are, they can't solve one problem: with a single line from you, AI fills in the blanks using its default aesthetic. A line from seeco's original post is spot-on: One line gets you Fable's taste. A director's brief gets you yours. A one-line prompt generates Fable's taste; a director's brief generates yours.
That means you need to translate feeling into numbers: use hexadecimal for background colors, Bezier curves for motion effects, and write the timeline in seconds. AI doesn't read “high-end”; it only reads roughness 0.38.
This workflow is currently very cheap. Someone used Fable 5.2 to make a 32-second nature-themed game video for a token cost of $37; a 3D game built from scratch cost about $550. Compared with outsourcing, that's almost negligible.
Another counterintuitive point: the most valuable items on this list aren't the generation tools, but the inspection tools. When AI judges motion from static screenshots, it misses frozen frames. seeco's tutorial specifically emphasizes using freezedetect and frame-by-frame comparison images for inspection. Tools don't replace people; they force AI to expose the process to you.
Finally, the barrier to this workflow is lower than you'd think. You don't need After Effects, you don't need an animation background, and you don't even need to know how to code. What you need is Claude, a browser, and the ability to write requirements as numbers.
The list is ready—save it. Next time you make a video, write the brief first, then let AI do the work.
发布时间: 2026-11-20 13:13