Viz3d skill lets Claude Code build and self-correct 3D simulations
A Japanese marketer tested 'viz3d', an open-source skill for Claude Code, to build 3D applications. The tool, published by codepawl under an MIT license, allows users to create interactive 3D scenes by describing them in plain language. The author tested the skill on an old iMac using a fictional restaurant lunch rush as the subject. The test proceeded in three stages: a basic version using…
Key points
- viz3d is an MIT-licensed skill for Claude Code that builds 3D apps with self-inspection.
- The AI corrected visual and logic errors by analyzing its own screenshots during the build.
- The test produced a 30-second 1080p video of a restaurant simulation on an old iMac.
The skill operates by following a defined procedure: planning, building models, assembling the app, and self-inspecting. A key feature is the 'self-inspection' cycle, where the AI takes screenshots of its own output, measures performance, and corrects errors such as overlapping surfaces or unrealistic movements. In the simple version, the AI identified and fixed a logic bottleneck where trays filled the counter, stopping the simulation. In the Blender version, it created 13 types of equipment and 8 types of animated people. The final stage added realistic human bodies and specific tool-handling movements, with the AI correcting visual glitches like clothing clipping.
The test demonstrated that viz3d can produce functional 3D simulations with calculated metrics, camera controls, and video export capabilities. The author noted that the process is deterministic, meaning the same inputs yield the same results. The skill requires Node.js, Playwright, and ffmpeg, with Blender optional for higher fidelity. The entire process, including model creation and video generation, took approximately an hour for the Blender version.
A Complete Guide to the AI Skill 'viz3d': Testing How Much It Can Do with Claude Code and an Old iMac|ポポコム武田|飲食を20年見てきたマーケター
note.com · 30 October 2026
Loading the full article…
This text was published by note.com and written by ポポコム武田|飲食を20年見てきたマーケター. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Meta launches Muse agent app in Japan with free tier and limited features · 1 src
- Open-source coding agent OpenCode lets users pick models and run locally · 1 src
- Cathedral accesses Anthropic internal systems in 12 hours, report says · 1 src
- Anthropic adds Internet Computer connector for Claude’s app interactions · 1 src
- Claude automates Google Sites build with AI-to-AI instructions · 6 src
Comments
via GitHub Discussions