lcm-demo's Introduction

LCM-LoRA Realtime drawing demo

This is a demo script for real-time drawing of LCM-LoRA. By dragging the middle line drawing to the input, the result of the output can be used for the input. The demo uses the sd1.5 model and control_v11p_sd15_scribble.pth.

controlnet weights is here and here(sdxl)
LCM-LoRA weights is here

https://twitter.com/minux302/status/1726058625381376329

Setup

controlnet_workflow.py is based workflow_api.py generated by ComfyUI-to-Python-Extension, so you need to create ComfyUI env.
If you simply want to try LCM-LoRA, I recommend another repository.

$ python -m venv .venv
$ source venv/bin/activate
$ pip install gradio==3.48.0

$ git clone https://github.com/comfyanonymous/ComfyUI
$ cd ComfyUI
$ git checkout d9d8702d8dd2337c64610633f5df2dcd402379a8
$ pip install -r requirements.txt

$ pip install torchvision===0.16.1
# This code uses comfy_controlnet_preprocessors(https://github.com/Fannovel16/comfy_controlnet_preprocessors).
# You need to install comfy_controlnet_preprocessors via ComfyUI-Manager(
https://github.com/ltdrdata/ComfyUI-Manager)

# since library conflicts with gradio...
$ rm ComfyUI/custom_nodes/ComfyUI-Manager
$ cp controlnet_workflow.py ComfyUI

Put weights to `ComfyUI/models`` and write weight names to ui.py, like

pipeline = LCMControlnetPipeline(
    ckpt_name="v1-5-pruned.ckpt",
    lcm_lora_name="pytorch_lora_weights.safetensors",
    control_net_name="control_v11p_sd15_scribble.pth",
    negative_prompt="low quolity"
)

run