New ControlNets Enhance Stable Diffusion 3.5 Large Image Generation

```html

New ControlNets for Stable Diffusion 3.5 Large Expand Creative Possibilities

Stability AI has released three new ControlNets for Stable Diffusion 3.5 Large: Blur, Canny, and Depth. This expansion offers users more precise control over the image generation process and opens up new creative applications. The models are freely available under the Stability AI Community License for both commercial and non-commercial purposes, provided annual revenue is under $1 million USD.

The New ControlNets in Detail

ControlNets allow the image generation process of Stable Diffusion to be controlled and refined through additional inputs. The three new models each offer specific functionalities:

- Blur: Enables high-resolution upscaling up to 8K and 16K. Particularly useful for creating detailed images from low-resolution templates. - Canny: Uses Canny edge maps to influence the structure of generated images. Especially suitable for illustrations, but also adaptable to other styles. - Depth: Uses depth maps, created for example with DepthFM, to control image composition. Ideal for architectural visualizations, 3D texturing, and other applications that require precise control over image composition.

Usage and Availability

The model weights of the new ControlNets are available on Hugging Face, the code is published on GitHub. Integration into ComfyUI simplifies usage and provides users with an intuitive interface. For local use, Stability AI recommends ComfyUI or the standalone SD3.5 repository for programmatic application. Detailed instructions and example workflows can be found in the documentation and on the respective platforms.

Advantages and Application Examples

The new ControlNets offer a number of advantages for creatives and developers:

- More Precise Control: Controlling image generation via Blur, Canny, and Depth allows for finer tuning of results and reduces the need for time-consuming post-processing. - New Creative Possibilities: The expanded control options open new avenues for artistic experimentation and the implementation of complex image compositions. - Increased Efficiency: By precisely controlling the generation process, high-quality results can be achieved faster and more efficiently.

Application examples for the new ControlNets include:

- Creating high-resolution images from low-resolution templates (Blur) - Generating illustrations with precise contours (Canny) - Creating realistic architectural visualizations and 3D textures (Depth)

Future Developments

Stability AI plans to release further ControlNet models, including variants for Stable Diffusion 3.5 Medium (2B) and new control types. These extensions are intended to further improve the functionality of Stable Diffusion and offer users even more creative possibilities. Mindverse, as a German provider of AI-powered content solutions, is closely following these developments and plans to integrate the new ControlNets into its platform to provide its customers with the latest AI technologies.

Conclusion

The new ControlNets for Stable Diffusion 3.5 Large represent a significant advance in AI-powered image generation. They offer users more control, open up new creative possibilities, and contribute to increased efficiency. The freely available license underscores Stability AI's commitment to open source and allows a wide user base to benefit from the innovations.

Bibliographie: https://stability.ai/news/sd3-5-large-controlnets https://blog.comfy.org/sd3-5-large-controlnet/ https://huggingface.co/stabilityai/stable-diffusion-3.5-controlnets https://huggingface.co/stabilityai/stable-diffusion-3.5-large-controlnet-blur https://twitter.com/btibor91/status/1861506709010964492 https://alternativeto.net/news/2024/11/stable-diffusion-3-5-large-launches-three-new-free-controlnet-models-blur-canny-and-depth/ https://www.all-ai.de/news/top-news24/stable-diffusion-3-5-so-ver%C3%A4ndern-neue-controlnets-alles https://openart.ai/workflows/cgtips/new-blur-controlnet-models-for-sd-35-large/UYUuSvurArMsMkOJQuyg ```