AI Videos Look Cheap, Florence Has Opus 5.5 Read These 14 Free Repos

The cheap feel of AI videos often gets stuck in the production chain. The generative model is only one part. After the visuals are generated, subtitles, pacing, 3D, rendering, export, and self-checks—if any link is missing, it shows.

Florence posted on X with a very direct title: Your AI videos look cheap? Send Opus 5.5 these 14 repos. The idea is to throw the repositories below at coding-capable agents like Opus 5.5. According to the original post, they are all free.

In the main video, a pink block character stands in front of a board that reads skills and Video guidance for your agent. It corresponds to Remotion Skills, the official video-building guidance for agents.

These 14 repositories can be broken down into several tracks.

Rendering and compositing: Remotion writes videos with React; Remotion Skills gives agents official guidance; HyperFrames writes HTML and renders video directly, explicitly designed for agents; Three.js runs 3D in the browser; GSAP handles UI motion and kinetic typography; FFmpeg handles encoding, checking, and outputting MP4.

3D and scenes: Blender does modeling, rigging, and real 3D rendering; Blender MCP lets Claude control Blender scenes.

Audio and subtitles: faster-whisper generates subtitles from any voiceover; Chatterbox generates open-source voiceover from plain text.

Assets and generation: ComfyUI connects image and video generation pipelines; yt-dlp pulls reference clips and audio from the web.

Design and review: Impeccable is used to polish, critique, and review design; Chrome DevTools MCP lets agents see their own rendered results and fix issues before export.

Full list:

• Remotion Skills: official video-building guidance for agents http://github.com/remotion-dev/skills
• Blender: modeling, rigging, and real 3D rendering http://github.com/blender/blender
• faster-whisper: generates subtitles from any voiceover http://github.com/SYSTRAN/faster-whisper
• Chatterbox: generates open-source voiceover from plain text http://github.com/resemble-ai/chatterbox
• GSAP: UI motion and kinetic typography on the timeline http://github.com/greensock/GSAP
• Blender MCP: lets Claude control Blender scenes http://github.com/ahujasid/mcp-for-blender
• Chrome DevTools MCP: lets agents see their own renders and fix them before export http://github.com/ChromeDevTools/chrome-devtools-mcp
• FFmpeg: encoding, checking, and outputting MP4 http://github.com/FFmpeg/FFmpeg
• Remotion: compose and render videos with React http://github.com/remotion-dev/remotion
• Impeccable: polish, critique, and review design http://github.com/pbakaus/impeccable
• Three.js: 3D in the browser, rendered with the GPU http://github.com/mrdoob/three.js
• ComfyUI: image and video generation pipelines http://github.com/Comfy-Org/ComfyUI
• yt-dlp: pull reference clips and audio from the web http://github.com/yt-dlp/yt-dlp
• HyperFrames: write HTML, render video, designed for agents http://github.com/heygen-com/hyperframes

The last line of the original post is Editor, voice actor, 3D artist → $0. It sounds like a free replacement for an entire post-production team. In reality, the barrier hasn't disappeared: the agent has to be able to string these tools together and check its own output.

In the comments, someone tried a workflow and called the result slopcore. Steven Ochs said the results from Opus 5.5 surprised him.

Others take a lighter path. FakerFill said plain three.js can go a long way; his 4K music video is a single renderAt(t) function, written with Claude, without using Blender.

Looking at it, the number of repositories is not the point. Self-check links like Chrome DevTools MCP are more worth paying attention to. For AI videos to escape the cheap feel, first let the agent see what it has rendered, then talk about style. Free tools can save the bills for an editor, voice actor, and 3D artist, but whether they save rework time depends on the workflow.