Artificial intelligence developer Stability AI has created generative AI technology that can generate 3D images at lightning speed. While previously requiring powerful hardware and complex programming, Stable Fast 3D’s two billion-parameter model can now generate vibrant, photorealistic images from text or reference in just half a second, VentureBeat reports.
Image source: Stability.ai
According to the company, the technology marks a significant breakthrough in the field of 3D modeling, reducing image processing time. “This is 1,200 times faster than our previous Stable Video 3D model, released in March, which took up to 10 minutes to create just one 3D object,” Stability AI notes.
Stable Fast 3D promises to be a powerful tool for a variety of industries, including design, architecture, retail, virtual reality and game development. It is based on TripoSR technology, developed in collaboration with 3D modeling company Trip AI.
Image source: Stability.ai
A research paper published by Stability AI details the methods used by the new model to quickly reconstruct high-quality 3D meshes from conventional 2D images. The essence of the method is to increase the speed of generating the final result without losing quality. At the same time, the technology uses an advanced transformer neural network to create high-resolution volumetric images without a significant increase in computing power, which makes it possible to reduce artifacts and obtain more detailed 3D models.
Additionally, Stable Fast 3D takes an innovative approach to evaluating lighting and materials. The AI model can determine global values, such as roughness, liquid, or metal, using a probability method that improves image quality. The technology also allows you to combine multiple elements needed for a 3D image, including meshes, textures and material properties, into a compact, ready-to-use 3D asset.
Interestingly, Stability AI continues to actively push the boundaries of generative AI, moving from 2D to 4D. The company, which began its journey with generating images from text – Stable Diffusion, released Stable 3D in November 2023. In March of this year, it debuted Stable Video 3D technology with the ability to basicly pan the camera to view images and improved quality of 3D image generation. And just last week it announced Stable Video 4D technology, which adds a time dimension to the generation of short 3D videos.
The Stable Fast 3D model is available through Stability AI’s Stable Assistant chatbot, Stability AI API, and licensed from the Hugging Face research community.