Main features

• Ultra-fast generation: It only takes 0.5 seconds to generate a complete 3D model, which is much faster than previous models.
• High Quality: The generated 3D models are of high quality, with detailed meshes and materials, and realistic colors.

Flexibility

• Re-meshing: You can choose to re-mesh quadrilateral or triangle meshing, which only takes an additional 100–200 milliseconds to make the model more suitable for different needs.
• Multi-purpose: Suitable for various fields such as gaming, VR, retail, architecture and design.

Ease of use

• Simple operation: Users only need to upload a picture and the system will automatically generate a detailed 3D model.
• API and Chatbot: Users can easily use this model through Stability AI’s API and Stable Assistant chatbot and interact with it on augmented reality devices.

Compatibility

• Augmented Reality: The generated 3D models can be displayed and interacted with in augmented reality on devices that support WebXR.
• Multiple application scenarios: can quickly create static objects in games, 3D product displays in e-commerce, models in AR/VR, etc.

Performance Advantages

• Ultra-high speed: Each 3D model takes only 0.5 seconds to generate, which is much faster than the previous 10 minutes.
• Maintain high quality: The exported 3D models are of high quality with detailed textures, and additional material parameters and normal maps can be generated.

Stable Fast 3D builds on our previous work on TripoSR , but features a completely retrained model and significant improvements in the architecture. These improvements allow for explicit mesh generation and incorporate new techniques for fast textured mesh creation.

A technical report is here , highlighting how we achieved fast inference by reducing baked lighting and material parameters