After much anticipation, SD3 has finally been officially released. In this video, I will guide you step-by-step through the entire process, from selecting and downloading the model to generating images in ComfyUI.
As a ready-to-use base model, the current experience is quite average. With three clips, including T5, it requires about 10G of VRAM. Although it has significantly improved its understanding of prompts, its performance in generating human images is quite poor. However, this is just a base model, and having issues is normal. It's best to wait for the fine-tuned versions.
stable-diffusion-3-medium :huggingface.co...
22 авг 2024