v3.1.2 Update:
Added LoRa Stacker
Simplified Loader
More Aspect Ratios and Resolutions
I actually wanted to add the Super Resolution ControlNet, but the integration into ComfyUI isn't working quite right yet. So, this feature will come with the next update.
V3.1 Update:
The V3.1 update now supports the ComfyUI Cascade models.
Additionally, it now has a cascade resolution and aspect ratio node, thus improving the quality of life. Many thanks to "bellamiss" for the node.
The Cascade resolutions custom node is included in the .zip file, just copy the node into the /customnodes folder."
ComfyUi Cascade Checkpoints Link:
V3 Update:
In V2, there was a bug that became apparent when rendering several hundred images, which affected the efficient sampler.
Therefore, I replaced it with the vanilla sampler in V3.
Many thanks for the help with testing to "erosdiffusion" from the L2 Discord.
Furthermore, V3 now has a face detailer that can edit up to 5 different faces.
Optionally, for images with larger faces, the quality of eyes can be improved with the eye detailer.
Eye Yolo Detector Model Link:
More Details LoRa SD 1.5 Link:
Update v3d (Diffusers):
After trying out numerous samplers and configurations, I've decided to go back to using the unofficial DiffusersStableCascade sampler again because it just consistently produces better, more creative images. Feel free to try out the v3 version as well, maybe you'll find a sampler configuration that consistently works well.
Diffusers Stable Cascase ComfyUi Node
It's possible that you encounter errors during installation. Here is a solution approach.
https://github.com/huggingface/diffusers/issues/1619#issuecomment-1345604389
https://huggingface.co/stabilityai/stable-cascade/discussions/17/files
V2 Update:
Comfyui now supports native stable cascade, so the diffusers sampler from v1 is no longer needed. The workflow generates images with sd cascade and improves face details if needed with an sd1.5 model.
The following resolutions work well: 1024x1024 / 1024x1904 / 1024x3808. If you know any other resolutions that work well, please post them here.
I've included a short guide on where to download the models and which folders to copy them into in the red notes of the workflow.
Description
LoRa Support
FAQ
Comments (8)
Looking forward to trying this. SC is one of my favorite checkpoints and you're wfs are amazing. Thanks for this!
In the cascade node, you can set different resolutions. Why are there some very large resolutions like 4K UHD available as options and how to get these to work? The model I assume can do in and around 1024 x 1024 correct? When going to resolutions far beyond this, things go wild with body autonomy. Please can you explain this node to me?
Cascade can render 4K images if your GPU has enough V-RAM.
@denrakeiw Oh really? I have a 4090 so that's no problem but I am getting weird results using higher resolutions like extra limbs, stretched torsos, multiple heads as what you would get if you tried to use dimensions way above the trained images for 1.5 and XL. Strange
@gohan2091 Yeah, it doesn't work with all motifs. It's always a bit of trial and error. But it works significantly better than with SD1.5 or SDXL models.
Can I make a request? Could you add an upscaler to this workflow? The faces look great but the bodies have the airbrushed skin. It would be good to generate then blow up the resolution with an upscaler group. Is this something you could add?
I always use the Supir upscaler, but combining it into a workflow with Cascade doesn't make sense because it requires too much VRAM, and most people don't want to wait that long or have an RTX A6000 in their computer.
https://civitai.com/models/364115/supir-upscale-comfyui
Error occurred when executing StableCascade_CheckpointLoader //Inspire: 'model.diffusion_model.input_blocks.0.0.weight' File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 74, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Inspire-Pack\inspire\backend_support.py", line 464, in doit b_model, clip, b_vae = nodes.CheckpointLoaderSimple().load_checkpoint(ckpt_name=stage_b) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\nodes.py", line 516, in load_checkpoint out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings")) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 458, in load_checkpoint_guess_config model_config = model_detection.model_config_from_unet(sd, "model.diffusion_model.") ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\comfy\model_detection.py", line 194, in model_config_from_unet unet_config = detect_unet_config(state_dict, unet_key_prefix) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\ComfyUI_windows_portable\ComfyUI\comfy\model_detection.py", line 78, in detect_unet_config model_channels = state_dict['{}input_blocks.0.0.weight'.format(key_prefix)].shape[0] ~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^