Hyperlora ComfyUI Installation

This is a simple guide to setting up Hyperlora in ComfyUI. I’ve found that Hyperlora is the easiest way to do a faceswap with just a single reference image.

This guide works on my system. I haven’t tried it anywhere else so I’d like to know about any issues that you run into. If you have problems getting it to run, leave a comment on my YouTube video describing your problem and I’ll help you out.

Setup

This guide is based off of the source Hyperlora page.

• cd to your ComfyUI models directory.
• Create a directory called hyper_lora and cd into it.
• Download the Hyperlora model files, use one method below, Windows method is a simple copy and paste.
  • Windows (simple method)
    • Run this command in a cmd terminal: winget install bodaay.hfdownloader. This installs a HuggingFace downloader.
    • Run this command in a cmd terminal: hfdownloader -m bytedance-research/HyperLoRA hyper_lora
  • Manual method
    • Create a directory called hyper_lora
    • Download sdxl_hyper_id_lora_v1_edit and sdxl_hyper_id_lora_v1_fidelity folders and all their files to the hyper_lora directory.
• Copy preprocessor_config.json, config.json, and model.safetensors to the respective directories. These are the first three files shown in the tree below. Create the directories as shown in the tree view.

C:.
└───hyper_lora
    ├───clip_processor
    │   └───clip_vit_large_14_processor
    │           preprocessor_config.json -> https://huggingface.co/openai/clip-vit-large-patch14/raw/main/preprocessor_config.json
    │
    ├───clip_vit
    │   └───clip_vit_large_14
    │           config.json -> https://huggingface.co/openai/clip-vit-large-patch14/raw/main/config.json
    │           model.safetensors -> https://huggingface.co/openai/clip-vit-large-patch14/resolve/main/model.safetensors
    │
    └───hyper_lora
        │   .gitattributes (not needed, but will show up if you download the entire repo)
        │   README.md      (not needed, but will show up if you download the entire repo)
        │
        ├───assets
        │       overview.jpg (not needed, but will show up if you download the entire repo)
        │
        ├───sdxl_hyper_id_lora_v1_edit
        │       hyper_lora_modules.json
        │       hyper_lora_modules.safetensors
        │       id_projector.safetensors
        │       resampler.safetensors
        │
        └───sdxl_hyper_id_lora_v1_fidelity
                hyper_lora_modules.json
                hyper_lora_modules.safetensors
                id_projector.safetensors
                resampler.safetensors

• Verify that everything is in your hyper_lora matches what is shown above. On Windows you can use tree /f to generate a tree view for comparison.
• Install the Antelopev2 model
  • Download the zip file
    • This link was obtained here.
  • Extract the files from the zip file and place them as shown in the tree below. The insightface directory is a directory in your main ComfyUI models directory. Create it if necessary.

models\insightface
        └───models
            └───antelopev2
                    1k3d68.onnx
                    2d106det.onnx
                    genderage.onnx
                    glintr100.onnx
                    scrfd_10g_bnkps.onnx

• Download the workflows to a temp directory
  • git clone https://github.com/bytedance/ComfyUI-HyperLoRA/
  • If you are on Windows and don’t already have git
    • Run this command in a cmd window first winget install Microsoft.Git
• Download the RealVisXL model (optional)
  • The Hyperlora devs tested on this model so it is the most likely to work. You can experiment with other SDXL models.
    • Download the v4.0 (BakedVAE) model and place it in your models\checkpoints folder.
• Start ComfyUI
• Go to the temp directory from above and cd to assets
• Drag the HyperLoRA-T2I-FaceDetailer.json file into ComfyUI.
• Go to the ComfyUI Manager, click ‘Install Missing Custom Nodes’, download everything shown.
• Restart and reboot the server when prompted.
• Drag the id-02.png file into the Load Image node in the ComfyUI workflow.
• Click the Run button. You’ll probably get an error in the Load Checkpoint node. Change the node to select the RealVisXL model that you downloaded.
• Click Run again. It should run successfully.
• Compare your result to the result on this page. It should be very similar.
• That’s it, if you have any issues then leave a comment on the YouTube video and I can help you get going.

Watch the video