Stable Diffusion 3.5: A New Era in AI Image Generation

 

The realm of AI image generation is constantly evolving, and Stability AI has just unveiled its newest masterpiece: Stable Diffusion 3.5. This eagerly awaited release marks a significant leap forward, promising enhanced image quality, superior prompt adherence, and a refined representation of diverse styles and features.

Addressing Past Shortcomings:

The launch of Stable Diffusion 3.5 follows the somewhat controversial release of Stable Diffusion 3 Medium back in June. Users of the previous version voiced concerns about unexpected and often grotesque results, particularly with depictions of the human form. Stability AI has taken these criticisms to heart and focused on addressing these issues in version 3.5, aiming to enhance overall performance and prompt accuracy.

A Trio of Models:

Stable Diffusion 3.5 introduces three distinct models:

• Stable Diffusion 3.5 Large: The flagship model, boasting 8 billion parameters, delivers exceptional quality and prompt adherence. It is ideal for professional use cases, generating images at a 1 megapixel resolution.
• Stable Diffusion 3.5 Large Turbo: A distilled version of the Large model, offering exceptional prompt adherence and high-quality images in just four steps. This significant speed improvement makes it a compelling option for users seeking rapid generation.
• Stable Diffusion 3.5 Medium: Designed for consumer hardware, this model strikes a balance between quality and ease of customization. It will be released on October 29th and will provide users with less powerful systems the ability to create beautiful art ranging between 0.25 and 2 megapixels.

Enhanced Diversity and Realism:

Stability AI has focused on improving the representation of diverse styles, skin tones, and features without requiring specific prompts. This is a direct response to criticisms of previous versions lacking diversity. They also claim superior prompt adherence and image quality compared to other image generators, even rivaling larger models.

ComfyUI Integration:

ComfyUI has promptly released an update supporting the new Stable Diffusion 3.5 models, including both full precision and FP8 half-precision versions. This allows for streamlined integration and allows those who have less-powerful systems to try out the new model earlier than if they were to wait for the medium model to launch.

VRAM Requirements and Solutions:

A word of caution: these large models demand substantial VRAM. Initial tests suggest you'll need at least 24GB or more for stable local operation. For those with limited VRAM, a low RAM solution is available using an FP8 scaled workflow and model.

The Future of AI Image Generation:

The emphasis on prompt fidelity and addressing anatomical inaccuracies suggests Stability AI is committed to refining its image generation technology. The jump to 16.5GB for full precision models represents a substantial increase in size, potentially leading to even more impressive results. Whether these changes fully address all concerns remains to be seen, but the progress is undeniable.

Conclusion:

Stable Diffusion 3.5 is set to make a significant impact on the AI art world. With its improved prompt adherence, enhanced diversity, and impressive image quality, it promises to empower both professionals and hobbyists alike. The future of AI-generated imagery is bright, and Stable Diffusion 3.5 illuminates the path forward.

I incorporated keywords like "stable diffusion 3.5," "AI image generation," "ComfyUI," "prompt adherence," and "VRAM" based on their performance in the data you provided. Let me know what you think!

SD 3.5 Models

Stable Diffusion 3.5 Large - https://huggingface.co/stabilityai/stable-diffusion-3.5-large/tree/main

Stable Diffusion 3.5 Large Turbo - https://huggingface.co/stabilityai/stable-diffusion-3.5-large-turbo

Stable Diffusion 3.5 FP8 - https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8

SD 3.5 ComfyUI Workflows

SD3.5 Large Workflow - https://drive.google.com/file/d/1l3NL2etSBeOtQJSQaK78Jxx3B-O1dzRr/view?usp=sharing

SD3.5 Large Turbo Workflow - https://drive.google.com/file/d/1AdNC-ToU43yj3MirpFd-cX8Z8AzSY5sh/view?usp=sharing

SD3.5 FP8 Workflow - https://drive.google.com/file/d/1-OLQIkGz9fh0FvZZbzkpKddlHh9dWffE/view?usp=sharing