ComfyUI_examples
SD3 Examples
SD3.5
The first step is downloading the text encoder files if you don’t have them already from SD3, Flux or other models: (clip_l.safetensors, clip_g.safetensors and t5xxl) if you don’t have them already in your ComfyUI/models/clip/ folder. For the t5xxl I recommend t5xxl_fp16.safetensors if you have more than 32GB ram or t5xxl_fp8_e4m3fn_scaled.safetensors if you don’t.
The SD3.5 model family contains a large 8B model and a medium 2.5B model. The medium model will be faster and take less memory but might have less complex understanding of some concepts. I recommend downloading both and experimenting with how each of them respond to your prompts.
The sd3.5_large.safetensors and sd3.5_medium.safetensors files (pick the one you want and put it in your ComfyUI/models/checkpoints/ directory) do not contain text encoder/CLIP weights so you must load them separately to use that file just like in the following example:
To use the sd3.5_large_turbo.safetensors file (put it in your ComfyUI/models/checkpoints/ directory) you can use the above example and set steps to 4 and cfg to 1.2.
For convenience there is an easy to use all in one checkpoint file sd3.5_large_fp8_scaled.safetensors (put it in your ComfyUI/models/checkpoints/ directory) that can be used in the default workflow like any other checkpoint files. There is also one for SD3.5 medium: sd3.5_medium_incl_clips_t5xxlfp8scaled.safetensors
See this workflow for an example.
As a reminder you can save these image files and drag or load them into ComfyUI to get the workflow.
SD3.5 Controlnets
Stability has released some official SD3.5 controlnets that you can find here these files (sd3.5_large_controlnet_canny.safetensors, sd3.5_large_controlnet_depth.safetensors, sd3.5_large_controlnet_blur.safetensors) go in your ComfyUI/models/controlnet directory and are meant to be used with SD3.5 large.
See this workflow for an example with the canny (sd3.5_large_controlnet_canny.safetensors) controlnet:
