SD3 Medium is Here!
Key Takeaways:
• 2 Billion Parameters: Stability AI’s most advanced text-to-image model yet.
• Accessible: Perfect for consumer PCs, laptops, and enterprise GPUs.
• Open Licensing: Available under a non-commercial license and a low-cost Creator License. Contact us for large-scale commercial use.
Features:
• Photorealism: Overcomes common artifacts in hands and faces.
• Prompt Adherence: Handles complex prompts with spatial relationships and styles.
• Typography: Generates text without artifacts using Diffusion Transformer architecture.
• Resource-efficient: Runs on standard consumer GPUs with low VRAM footprint.
• Fine-Tuning: Customizable with small datasets.
Collaborations:
• NVIDIA: Enhanced performance with RTX GPUs and TensorRT for a 50% boost.(
here)
• AMD: Optimized inference for various AMD devices.
Open and Accessible:
• Released under the Stability Non-Commercial Research Community License.
• Professional use encouraged with our Creator License. Contact us for commercial licensing.
(TLDR generations are still free to use, commercial and donate-ware finetunes are paid now)
Safety:
• Committed to safe, responsible AI with extensive testing and safeguards.
(read as lobotomized in the same way as SDXL was)