* Unlike the Stable Diffusion 3.0 release, where the public release was initially (and ultimately only) the Medium model, this is both the base and distilled (fewer steps needed) Turbo versions of the Large model (8B).
* ComfyUI has immediate support for SD3.5 Large [0], and comfy.org has released an FP8 scaled version of the model [1] for lower memory use (note that the FP8 version is almost as large in file size as the base FP16 model, but that's because it has the 3 text encoders baked in, while the base model relies on them being distributed separately.)
Some notes/resources:
* Unlike the Stable Diffusion 3.0 release, where the public release was initially (and ultimately only) the Medium model, this is both the base and distilled (fewer steps needed) Turbo versions of the Large model (8B).
* ComfyUI has immediate support for SD3.5 Large [0], and comfy.org has released an FP8 scaled version of the model [1] for lower memory use (note that the FP8 version is almost as large in file size as the base FP16 model, but that's because it has the 3 text encoders baked in, while the base model relies on them being distributed separately.)
* Even smaller bitsandbytes nf4 version [2]
* Some SD3.5 LoRAs already released. [3][4][5]
* Fine-tuning guide. [6]
[0] https://blog.comfy.org/sd3-5-comfyui/
[1] https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/tr...
[2] https://huggingface.co/sayakpaul/sd35-large-nf4
[3] https://huggingface.co/Shakker-Labs/SD3.5-LoRA-Chinese-Line-...
[4] https://huggingface.co/Shakker-Labs/SD3.5-LoRA-Linear-Red-Li...
[5] https://huggingface.co/Shakker-Labs/SD3.5-LoRA-Futuristic-Bz... (don't ask me why it is "bzonze" and not "bronze".)
[6] https://stabilityai.notion.site/Stable-Diffusion-3-5-Large-F...