LCM_outpaint_promptless.json issue

[danzelus]

My comfy (portable) path: B:\!Comfyui\ComfyUI My cnet path: B:\!Comfyui\ComfyUI\models\controlnet

win 11

[taabata]

For the loader model path, its for the LCM model not controlnet model. The model is automatically loaded once placed in models/diffusers folder (look at the readme for more details) or you can paste the model's full path where it says model_path in the loader node. On another note, this node is only for inpainting with reference and no controlnet. To use controlnet, there is another node 'LCMGenerate_inpaintv2', for that you can inpaint with controlnet. keep in mind you need to have the controlnet model with its config file in a folder inside the models/controlnet folder in your ComfyUI directory (model should be a folder containing the diffusion_pytorch_model.bin and the config.json). Also, at the moment you need to have a T2I adapter also connected to the node. I suggest using the workflows provided since the nodes and their inputs are not organized at the moment due to me being lazy. Sorry for any inconveniences.

[danzelus]

I tried to update diffusers (pip install --upgrade diffusers[torch], pip install --upgrade diffusers[flax], pip install diffusers==0.23.0) but still not working

[cardenluo]

the same issue ,Can you give me some suggestions.thanks

[taabata]

the same issue ,Can you give me some suggestions.thanks

I suggest you download all of the files and folders found here https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7/tree/main (not the .safetensors file) and place them inside a folder named 'LCM_Dreamshaper_v7' in the same directory you have the safetensors file at right now. Download this file https://github.com/taabata/LCM_Inpaint_Outpaint_Comfy/blob/main/preprocessor_config.json and place it in the folder as well.

[alexanderdutton]

Same issue.

I think I got everything in the right place. I tried manually filling in the model_path with the diffusers folder, the controlnet folder, and every other possible thing I could think of, but couldn't get the 'controlnet_model' to display anything other than 'cn_canny' when it first loads, and then 'undefined' when I try to refresh/cycle the options.

Fully perplexed.

Quite looking forward to working with this tool -- thanks for any help you might provide. :)

[KimMatt]

After reading the readme it's unclear to me what you mean by "put the diffusers version in" I see a lot of config.jsons for the different folders on huggingface (vae, vae_decoder, etc.) but am unsure which one of those you mean to put in?

[alexanderdutton]

It wants the entire hugging face diffusers folder in that subdirectory, as in clone the full folder, not simply grab a particular set of files.

But even after I did that, I was running into other errors, which I recall seemed to be because the system has hard filename expectations instead of using modular inputs.

[taabata]

You can use the newer inpaint nodes using LCM Lora. Instructions:

Clone the github repository into the custom_nodes folder in your ComfyUI directory

Run the setup script for the CanvasTool

Install any sd 1.5 based model in format that runs with diffusers (like this one https://huggingface.co/stablediffusionapi/deliberate-v2/tree/main) and place in models/diffusers folder in your ComfyUI directory

Install LCM_lora from https://huggingface.co/latent-consistency/lcm-lora-sdv1-5/tree/main and place in models/loras folder in your ComfyUI directory

Install controlnet inpaint model in format that runs with diffusers (from https://huggingface.co/lllyasviel/control_v11p_sd15_inpaint/tree/main) and place in models/controlnet folder in your ComfyUI directory

Install ip adapter models and image encoder and place in models/controlnet/IPAdapter (you have to create the folder) in your ComfyUI directory (optional; can use reference only instead)

Open the workflow in ComfyUI (https://github.com/taabata/LCM_Inpaint_Outpaint_Comfy/blob/main/inpaint_LCMLORA_promptless.json)

Also for the 'Get image size' node, you need to git clone https://github.com/BadCafeCode/masquerade-nodes-comfyui in your custom_nodes folder

Image resize node from https://github.com/WASasquatch/was-node-suite-comfyui

**meaning of 'in format that runs with diffusers': download the model's different component in their separate folders (vae,text encoder, unet, etc)