Stable Diffusion 3 Installation Guide

In the latest video, Olivio Sarikas dives into the new Stable Diffusion 3 (SD3) Medium Model, providing a comprehensive guide on how to install and run it locally on your computer. He starts by showcasing some images generated with SD3, emphasizing that these are first-roll outputs without any prompt optimization. The video promises further testing and quality improvement tips in a follow-up video. To begin, users are directed to the Hugging Face website to sign a free, non-commercial license agreement. For commercial use, users must contact Stability AI for a commercial license. Sarikas advises downloading the ‘sd3 medium including clip save tensor’ model, which is approximately 6 GB, as opposed to the ‘sd3 medium safe tensor’ model, which lacks a text encoder. He also mentions a larger 11 GB model that includes clip and T5 XXL fp8, which users might try if their systems can handle it. The downloaded model should be placed in the models folder of either Automatic 1111 or ComfyUI. Sarikas prefers using ComfyUI, noting that it often receives updates first. He explains how to update ComfyUI and its custom nodes, cautioning that an update might break the Torch Cuda model, which can be fixed by running an update script in the ComfyUI Windows portable folder. Once updated, users can load workflows from Comfy Anonymous, the developer of ComfyUI. These workflows include basic, multi-prompt, and upscaling workflows, which can be downloaded and tested within ComfyUI. Sarikas demonstrates a workflow setup for the ‘sd3 medium including clip save tensor’ model, showing how to load the checkpoint, write prompts, and adjust settings. He uses a prompt ‘cat holding a sign with the text I love you,’ which interestingly generates an image with a heart, showcasing the model’s creative interpretation. The video concludes with Sarikas encouraging viewers to like, subscribe, and stay tuned for more content.