Stability AI has released an AI 3D model and image creator called Stable Zero123, which can turn 2D images into 3D content. This creator is still under development but has already attracted the attention of many creators and developers, especially those in the video and gaming industries.

Stable Zero123 uses a method called Score Distillation Sampling (SDS) to analyze and reconstruct the depth and dimensions of objects from a single photo. This is particularly helpful for virtual reality and 3D design, such as engineering and architecture.

Stable Zero123 is provided through the Hugging Face platform, a place for sharing machine learning models. Stability AI recommends using Stable Zero123 together with the Three Studio software to better manage 3D content.

Stability AI is also developing other tools that can be used in conjunction with Stable Zero123 to enhance the functionality of the models. These tools include a sky replacer and a 3D model maker, both of which are currently in private preview. These tools provide users with more choices and effects.

While Stable Zero123 is powerful, it does have certain requirements that may pose difficulties for some users. The AI model requires strong computational power, necessitating advanced graphics cards or specialized training GPUs to fully utilize. This may limit the accessibility of the model, especially for enthusiasts or small creators without these resources.

The technical details and innovations of Stable Zero123 are as follows:

• Based on Stable Diffusion 1.5.
• Generates a new perspective with VRAM and SD1.5.
• Generating 3D objects requires more time and memory (recommended 24GB VRAM).
• For research purposes only, not for commercial use.
• Weights can be downloaded.
• Uses high-quality 3D object dataset from Objaverse.
• Utilizes highly conditional training and inference to improve prediction quality.
• Employs a precomputed dataset and improved data loader, resulting in a 40x training efficiency improvement.
• Released on Hugging Face for researchers and non-commercial users.
• Improved open-source code of threestudio to support Zero123 and Stable Zero123.
• Optimized a NeRF combined with Stable Zero123 using SDS.
• Capable of generating text-to-3D.
• Contact information available for inquiries regarding commercial applications.
• Offers newsletters, social media, and a Discord community for updates and more information.

One current drawback of Stable Zero123 is its inability to generate images with transparent backgrounds, which may affect its effectiveness in videos. However, it holds great potential in the video and gaming industries, where high-quality 3D content is in demand.

Stability AI continues to improve the application of Stable Zero123 and overcome its limitations. They also provide courses on machine learning and stable diffusion to help creators learn the knowledge and tools necessary for successful creative projects.

Stable Zero123 is a milestone in the AI-driven 3D image field. While still in development, it has already made a significant impact on content creation. Stability AI will continue to refine and enhance this technology, providing more advanced and accessible tools for creators and developers. The future of Stable Zero123 is highly anticipated, and the creative community is excited about the new possibilities that Stability AI brings to digital content creation.