RODIN Diffusion

RODIN Diffusion is a research neural network from a team associated with Microsoft. It uses a diffusion-based approach to generate 3D avatars from a portrait photo or from a text description. The official site does not provide a public interface or “try it” button, so it’s best understood as a research demo rather than a consumer-ready service.

How it works

The model builds a 3D figure in multiple stages:

• A rough pass that establishes core body and facial structure
• Refinement that adds volume, textures, and lighting

This staged pipeline helps preserve appearance details even when the input is just a single photo.

Generation modes and editing

RODIN Diffusion supports several workflows:

• Reconstructing an avatar from one photo
• Generating a figure from text prompts (e.g., hair color, clothing details, facial traits)
• Editing an existing generated figure (e.g., changing hairstyle or adding accessories)

Limitations and safety notes

There is no open/public version available, so you can’t upload a photo and receive a personal 3D avatar through an online tool. The project shares research materials and examples only. The authors also note the risk of misuse for fakes and recommend labeling generated results to reduce fraud and misinformation.