/**/ THIS DOCKER TAKES A LONG TIME TO BOOT THE FIRST TIME SEE LOG FOR OUTPUT /**/ Streamlit ui for Stable Diffusion: https://github.com/sd-webui/stable-diffusion-webui/ This now includes all models: SD v1-4 (ckpt) Waifu Diffusion (ckpt) TrinArt (ckpt) ViT-L-14 (pytorch.bin, used by inference models, is usually ~1.8gb autodownload to ~/.cache/huggingface/transformers) SD v1-4 (diffusers, used by txt2vid and textual inversion training) BLIP CLIP models: ViT-L-14, ViT-H-14, ViT-g-14, used by img2txt sd-concepts-library: (nearly) 600 pre-trained concepts to use in prompts, autoloaded when the placeholder is used in the prompt. GFPGAN v1.4 (and other weights that would normally be autodownloaded, detection_Resnet50_Final.pth and parsing_parsenet.pth, ~185MB) RealESRGAN: RealESRGAN_x4plus and RealESRGAN_x4plus_anime_6B Latent Diffusion Super Resolution (LDSR) Notes: Most models will unload when you use something else, atm there is a config option for img2txt.keep_all_models_loaded: False if you set this to True img2txt models (the CLIP ones, BLIP stays loaded unless you have optimized mode available) will all stay loaded, by default if a model that is loaded is no longer selected it will be unloaded volume path now mirrors the repo so is /sd/outputs instead of /sd/user_data/outputs