Stability AI Open-Sources SD3 Medium Model 206

Stability AI has just open-sourced its Stable Diffusion 3 Medium model (abbreviated as SD3), which is the latest and most advanced text-to-image AI model in the Stability AI Stable Diffusion 3 series. SD3 can run on consumer-grade laptops, and the SD3 model contains 2 billion parameters. It shows a significant improvement in image quality, capable of generating higher-quality and more detailed images. It can convert text descriptions into images more accurately.

The smaller size of the Stable Diffusion 3 Medium model allows it to run well on consumer-grade PCs and laptops as well as enterprise-grade GPUs. At the same time, this size characteristic also gives it the potential to become the next standard for text-to-image models.

Features of SD3 Medium:

• Improved image quality, generating higher-quality and more detailed images.

• Complex prompt understanding, capable of converting complex text descriptions into images more accurately.

• Overcomes common hand and facial flaws, providing high-quality images without complex workflows.

• Leveraging our Diffusion Transformer architecture, it achieves unprecedented text quality, reducing errors in spelling, kerning, letter formation, and spacing.
  

• Better at understanding and handling prompts involving spatial relationships, compositional elements, actions, and styles.

• Achieves unprecedented results in text generation, without manual artifacts or spelling errors.

• Achieves high performance with fewer computational resources, especially suitable for standard consumer-grade GPUs with low VRAM usage.

• Supports fine-tuning, ideal for customization.

Key Points of SD3 Medium:

• Stable Diffusion 3 Medium is Stability AI's most advanced text-to-image open model to date.

• This model is small in size, making it ideal for running on consumer PCs and laptops as well as enterprise-grade GPUs. Its appropriate size can make it the next standard for text-to-image models.

• These weights are now available under an open non-commercial license and a low-cost creator license. For large-scale commercial use, please contact us for licensing details.

• To try the Stable Diffusion 3 model, please try using the API on the Stability Platform, sign up for a three-day free trial on Stable Assistant, and then try Stable Artisan through Discord.

The SD3 Medium model was pre-trained on 1 billion images, including synthetic data and filtered public data. Fine-tuning data includes 30 million high-quality aesthetic images and 3 million preference data images to focus on specific visual content and styles.

SD3 Medium Official Introduction: https://stability.ai/news/stable-diffusion-3-medium

SD3 Medium Model Download: https://huggingface.co/stabilityai/stable-diffusion-3-medium

SD3 Medium Base Workflow: https://github.com/ZHO-ZHO-ZHO/ComfyUI-Workflows-ZHO