Stability AI Accelerates 3D Image Generation with Stable Fast 3D Technology

Not long ago, generating 3D images was a difficult and time-consuming task, requiring complex wireframes, specialized software, and powerful hardware. Today, that has changed dramatically.

Stability AI has introduced a groundbreaking generative AI technology called Stable Fast 3D, which enables the rapid creation of 3D images from a single picture in just half a second. This represents a significant advancement in processing time, offering results 1200 times faster than previous models, such as the Stable Video 3D (SV3D), which could take up to 10 minutes for similar outputs.

The implications of Stable Fast 3D are vast, with potential applications in design, architecture, retail, virtual reality, and game development. You can access this model through Stability AI’s Stable Assistant chatbot and the Stability AI API, as well as via a community license on Hugging Face.

The Technology Behind Stable Fast 3D

Stable Fast 3D evolves from Stability AI's prior work with the TripoSR model. In March, the company partnered with Trip AI to develop fast 3D asset generation technology.

In their research paper, Stability AI's researchers describe the innovative methods employed to swiftly reconstruct high-quality 3D meshes from single images. By integrating several novel techniques, they tackle common challenges in quick 3D reconstruction while enhancing both speed and output quality.

At its core, Stable Fast 3D utilizes an advanced transformer network that generates high-resolution triplanes—3D volumetric representations—directly from the input image. This network efficiently manages larger resolutions without significantly increasing computational demands, capturing finer details and minimizing aliasing artifacts.

Additionally, the model features a unique approach to estimating materials and lighting. Through a novel probabilistic method, the material estimation network predicts global metallic and roughness values, leading to improved image quality and consistency. Notably, Stable Fast 3D also integrates essential components for a complete 3D image—mesh, textures, and material properties—into a streamlined, ready-to-use asset.

Stability AI's Ongoing Innovations

Stability AI is widely recognized for its Stable Diffusion text-to-image generation technology. While Stable Diffusion focuses on 2D images, the company has been advancing its 3D capabilities since November 2023, beginning with Stable 3D. The subsequent release of Stable Video 3D in March enhanced 3D image generation quality and introduced basic camera panning for image viewing.

The company is not stopping at 3D; they recently unveiled Stable Video 4D, which incorporates time into short 3D video generation, pushing the boundaries of generative AI even further.