Evening news on June 12: AI startup Stability AI announced the official open-source release of its latest text-to-image generation model, Stable Diffusion 3 Medium (SD3 Medium). Stable Diffusion 3 Medium contains 2 billion parameters and is Stability AI's most advanced text-to-image open model to date. Its smaller VRAM footprint is intended to make it better suited for running on consumer-grade GPUs as well as enterprise-grade GPUs.
Hugging Face model address:https://huggingface.co/stabilityai/stable-diffusion-3-medium
Basic information about Stable Diffusion 3 Medium
- Advancedness: SD3 Medium has 2 billion parameters and is Stability AI's most complex image generation model to date, representing an important milestone in the evolution of generative AI.
- Applicability: This model's moderate size makes it an ideal choice for running on consumer PCs, laptops, and enterprise-grade GPUs, and it is expected to become a new standard for text-to-image models.
- Openness: The weights of SD3 Medium are now available under a non-commercial license and a low-cost creator license, encouraging professional artists, designers, developers, and AI enthusiasts to make commercial use of it.
- Multi-platform support: API trials are supported, and it can be tried on the Stability platform,Stable Assistant(free three-day trial), and via Discord onStable Artisan.
Stable Diffusion 3 Medium'simprovements
- Photorealism: SD3 Medium reduces common artifacts in hands and faces, delivering high-quality images that can be generated without complex workflows.
- Prompt adherence: The model can understand and generate complex prompts involving spatial relationships, compositional elements, actions, and styles.
- Typesetting capability: With the Diffusion Transformer architecture, SD3 Medium achieves artifact-free and typo-free text generation.
- Resource efficiency: The model is designed with resource efficiency in mind, maintaining high performance even on standard consumer-grade GPUs without sacrificing performance.
- Fine-tuning capability: SD3 Medium can absorb details from small datasets, providing strong support for customization.
Partners
- NVIDIA: Stability AI's collaboration with NVIDIA provides significant performance improvements for all Stable Diffusion models, including SD3 Medium, through NVIDIA® RTX™ GPUs and TensorRT™.
- AMD: AMD has optimized SD3 Medium, ensuring efficient inference on AMD's latest APUs, consumer-grade GPUs, and MI-300X enterprise-grade GPUs.
Openness and safety
- Stability AI is firmly committed to open generative AI. SD3 Medium is released under the Stability NonCommercial Research Community License, and a new Creator License has also been introduced to support commercial use.
- The company has taken comprehensive safety measures, from model training to deployment, to ensure the safe and responsible use of SD3 Medium.
Future plans
Stability AI plans to continuously improve SD3 Medium based on user feedback, expand its capabilities, and improve performance. The company's goal is to set a new standard for AI-generated art, making SD3 Medium an important tool for professionals and enthusiasts.
(Source:Stability AI）