SD3 - Local Install Guide! FASTEST Way to run the new Model

The video introduces Stable Diffusion 3 Medium (SD3) and demonstrates how to install and run the new model using ComVEI. Viewers are directed to the Hugging Face website to download the model, set up ComVEI, and access various workflows for different tasks, with a demonstration of running a Tex to Image workflow using a specific prompt.

In the video, the presenter introduces Stable Diffusion 3 Medium and shares some rendered images using the software. They mention that the images are unaltered and will conduct testing to showcase image quality improvements in the next video. To get started with SD3, viewers are directed to the Hugging Face website to sign a free license for non-commercial use. Different versions of the model are available, with the presenter recommending the SD3 Medium including Clip Save Tensor file.

To download the model, viewers are instructed to choose the appropriate version and save it in the models folder. The presenter mentions the availability of workflows for different tasks within ComVEI, such as basic, multi-prompt, and upscaling workflows. Additionally, a demo prompts file is provided for testing the model using various prompts. The video then focuses on setting up ComVEI to run the SD3 model, emphasizing the need to update ComVEI before using the model. In case of any issues, viewers are advised to update ComVEI and Python dependencies from the portable folder.

Further instructions are provided on how to load workflows in ComVEI, with links to workflows created by ComVEI Anonymous for different model versions. The presenter demonstrates loading a workflow for the SD3 Medium model with Clip Save Tensor and provides insights on settings suggested by ComVEI Anonymous, such as using the sgm uniform scheduler with specific values. A Tex to Image workflow is showcased with a prompt of ‘cat holding a sign with the text I love you,’ resulting in an image that creatively interprets the prompt. The video concludes with a thank you message, encouraging viewers to like, subscribe, and stay tuned for more content.