Customize Stable Diffusion LoRas

The ComfyUI StreamDiffusion custom node supports SD 1.5 models and LoRa checkpoints. LoRas are used to generate a variety of special effects in your workflow, such as color correction, sharpening, and more.

​

Configure LoRa checkpoint using ComfyUI

ComfyUI will automatically generate a workflow json file from changes in the LightGraph UI editor. Follow these steps to customize the LoRa checkpoints in your Stream Diffusion workflow:

1. Launch ComfyUI with the StreamDiffusion node installed
2. Choose an example Stream Diffusion workflow file that uses LoRas
3. Open the workflow file in ComfyUI
4. Change the lora_name input in the workflow to reflect the name of a LoRa checkpoint file you want to use
5. From ComfyUI, run the workflow to validate the change, save the workflow file when done: File > Save
6. Export and download the new workflow json file from ComfyUI: File > Export > API

​

Checkpoint Files

• LoRa checkpoints should be downloaded to ComfyUI/models/loras

  • Deployed LoRas are available at Livepeer-Studio/comfystream_loras

• SD 1.5 model checkpoints should be downloaded to ComfyUI/models/checkpoints

  • Deployed SD 1.5 models are available at Livepeer-Studio/comfystream_checkpoints

​

Configure LoRa checkpoint in workflow manually

​

Changing the LoRa checkpoint

The lora_name input will reflect the name of a LoRa file from Livepeer-Studio/comfystream_loras:

  "181": {
    "inputs": {
      "lora_name": "ral-alienzkin-sd15.safetensors"
    },
    "class_type": "StreamDiffusionLoraLoader"
  },

Update the lora_name input to the name of the LoRa checkpoint you want to use.

​

Changing the StreamDiffusion model

LoRas are only compatible with specific SD models. If you need to change the model, the checkpoint input defines the stable diffusion model (sd_turbo, sd1.5, or a fine-tuned variant) that will be loaded by the workflow:

  "178": {
    "inputs": {
      "checkpoint": "lahmixMysterious_v20.safetensors"
    },
    "class_type": "StreamDiffusionCheckpointLoader"
  },

​

Submit your LoRa Checkpoint to Livepeer

To request your LoRa checkpoint to be officially deployed on the Livepeer Network:

• Open a pull request to the hugging-face repository https://huggingface.co/Livepeer-Studio/comfystream_loras
• Open a Pipeline Request ticket with the Livepeer AI Support Team.