Stable Diffusion 3 Medium: The New Frontier in local AI Image Generation | How to use Dalle 3 in Bing | StabilityAI | Stable Diffusion | Turtles AI

Stable Diffusion 3 Medium: The New Frontier in local AI Image Generation
Stability AI launches a 2 billion parameter image generation model, optimized for consumer GPUs with flexible licensing.
DukeRem

Stability AI (finally) releases Stable Diffusion 3 Medium: a 2 billion parameter image generation model.

Stability AI has just announced the release of Stable Diffusion 3 Medium, a sophisticated version of their text-to-image generation technology. This model, with 2 billion parameters, represents a significant advancement in image generation, combining advanced diffusion transformer techniques and flow matching to enhance the quality and accuracy of the generated images.

Stable Diffusion 3 Medium excels at handling complex prompts involving spatial relationships, compositional elements, and various artistic styles. Additionally, the model shows improvements in generating photorealistic images, overcoming common representation flaws in hands and faces. Another strength is its ability to generate text within images without spelling errors, thanks to its advanced architecture.

From a technical standpoint, SD3 Medium uses a Diffusion Transformer Architecture that optimizes image generation, making it more faithful and precise to the initial prompts. This model is designed to be resource-efficient, operating smoothly on standard consumer GPUs due to its low VRAM footprint.

Stability AI has collaborated with NVIDIA to optimize the performance of all Stable Diffusion 3 models, leveraging RTX GPUs and TensorRT to boost performance by 50%. Simultaneously, AMD has enhanced SD3 Medium’s inference for various AMD devices, including the latest APUs, consumer GPUs, and MI-300X enterprise GPUs.

Stable Diffusion 3 Medium is released under an open non-commercial license and a low-cost Creator License. For large-scale commercial use, Stability AI requires direct contact for licensing details. The community can access this model via the API on the Stability platform, with a free three-day trial available on Stable Assistant and Stable Artisan via Discord.

Technical Features of Stable Diffusion 3 Medium

Feature Description
Parameters 2 billion
Architecture Diffusion Transformer
Collaborations NVIDIA (RTX, TensorRT), AMD (APU, GPU, MI-300X)
Licenses Non-commercial, Creator License, Enterprise License on request

 

Stability AI’s commitment to safe generative AI is evident from the numerous safety measures implemented during the model’s training and release. Extensive internal and external testing has been conducted to prevent misuse by malicious actors.

Highlights:

  • Stable Diffusion 3 Medium uses 2 billion parameters for advanced image generation.
  • The model is optimized for consumer GPUs, thanks to low VRAM requirements.
  • Collaborates with NVIDIA and AMD to optimize performance.
  • Released under non-commercial and commercial licenses, available via API with a free trial.