SEEAvatar

- Powered by large-scale text-to-image generation models, text-to-3D avatar generation has made promising progress. However, most methods fail to produce photorealistic results, limited by imprecise geometry and low-quality appearance. Towards more practical avatar generation, we present SEEAvatar, a method for generating photorealistic 3D avatars from text with SElf-Evolving constraints for decoupled geometry and appearance.
- For geometry, we propose to constrain the optimized avatar in a decent global shape with a template avatar. The template avatar is initialized with human prior and can be updated by the optimized avatar periodically as an evolving template, which enables more flexible shape generation. Besides, the geometry is also constrained by the static human prior in local parts like face and hands to maintain the delicate structures.
- For appearance generation, we use diffusion model enhanced by prompt engineering to guide a physically based rendering pipeline to generate realistic textures. The lightness constraint is applied on the albedo texture to suppress incorrect lighting effect.
- Experiments show that our method outperforms previous methods on both global and local geometry and appearance quality by a large margin. Since our method can produce high-quality meshes and textures, such assets can be directly applied in classic graphics pipeline for realistic rendering under any lighting condition.

All avatars are generated and rendered in full-body manner. Heads are rendered in a larger size for better view.


Batman	Joker	Spiderman


Abraham Lincoln	Barack Obama	Hillary Clinton


Ludwig van Beethoven	Wolfgang Amadeus Mozart	Vincent van Gogh


Captain Jack Sparrow	Doctor Strange	Elsa in Frozen

Comparison with text-to-avatar methods, NeRF-based DreamWaltz and mesh-based TADA.

Generated avatars are animated by retargeting motion sequences from Mixamo.

Generated avatars are lighted by multiple lights with different colors from different angles.

Photorealistic avatars represented in meshes and textures generated from text prompt by our method, exported to Blender for offline rendering, with additional 3D assets for building scenes.

Photorealistic avatars represented in meshes and textures generated from text prompt by our method, exported to Unreal Engine for real-time rendering, with additional 3D assets for building scenes.