hy-world-2-0-3d-world-model — Claude Code skill: 3D from photos
A skill for running HY-World 2.0: it turns photos and video into 3D scenes and builds worlds from text or a single image.
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a claude-codeWhat hy-world-2-0-3d-world-model does
This skill covers running Tencent Hunyuan's HY-World 2.0 model inside your own project. Two paths are documented. Reconstruction takes a folder of images or a video and returns a point cloud, depth maps, surface normals and 3D Gaussian Splatting attributes. Generation goes the other way: text or one picture becomes a full 3D world through four stages, from panorama to final composition.
WorldMirror 2.0 handles reconstruction. It is a feed-forward model of roughly 1.2B parameters that predicts depth, normals, camera parameters and 3DGS in one pass. Weights arrive from Hugging Face on the first run, but the environment is on you: Python 3.10, PyTorch 2.4.0, CUDA 12.4. Get the CUDA version wrong and the PyTorch install fails, so check your driver first.
Known camera parameters or a depth map can be fed into the pipeline as priors, which sharpens the result. Resolution is adjustable, 512 to 1024 on the shorter edge, and the low setting is the fast one. A Gradio app serves an interactive UI at http://localhost:7860.
What it helps with
- build a 3D scene from a folder of photos
- pull frames out of a video and reconstruct a point cloud
- generate a 3D world from a text prompt
- inject known camera parameters to sharpen reconstruction
- batch several scenes and save each result to its own folder
Example prompts for the agent
- Reconstruct a 3D scene from the images in
images/and save it tooutput_3d/ - Extract frames from
scene.mp4at 2 fps and run WorldMirror 2.0 on them - Generate a 3D world from a text description and export the mesh and 3DGS
How to install hy-world-2-0-3d-world-model
1Skills CLI (Vercel)
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a claude-code2Manually: put the files in place
~/.claude/skills/hy-world-2-0-3d-world-model/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/hy-world-2-0-3d-world-model/
1Skills CLI (Vercel)
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a codex2Manually: put the files in place
.agents/skills/hy-world-2-0-3d-world-model/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a cursor2Manually: put the files in place
.cursor/skills/hy-world-2-0-3d-world-model/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install reason-machines/trending-skills hy-world-2-0-3d-world-model2Manually: put the files in place
.github/skills/hy-world-2-0-3d-world-model/SKILL.md1Skills CLI (Vercel)
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a gemini-cli2Platform CLI
gemini skills install https://github.com/reason-machines/trending-skills --path skills/hy-world-2-0-3d-world-modelManually: put the files in place
.devin/skills/hy-world-2-0-3d-world-model/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about hy-world-2-0-3d-world-model
How do I install hy-world-2-0-3d-world-model in Claude Code?
Run npx skills add reason-machines/trending-skills --skill hy-world-2-0-3d-world-model -a claude-code. The skill lands at ~/.claude/skills/hy-world-2-0-3d-world-model/SKILL.md.
What does hy-world-2-0-3d-world-model do?
It documents running HY-World 2.0: reconstructing 3D from images or video, and generating a world from text or a single image. Outputs include a point cloud, depth maps, normals, camera parameters and 3DGS attributes.
What do I need to run it?
Python 3.10, PyTorch 2.4.0 and CUDA 12.4. Model weights download from Hugging Face on first run, and you can point the cache elsewhere with HF_HOME.
