BFS: Best Face Swap

BFS: Best Face Swap

Swaps faces, heads or bodies in images with LoRA files for Qwen Image, Flux 2 Klein, Krea 2 and LTX-2 in ComfyUI. Use it only with faces and people you have the right to use.

Made by Alissonerdx on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: BFS is a collection of LoRA files for replacing a face, head or body in an image. It covers several base models: Flux 2 Klein, Krea 2, Qwen Image Edit (2509 and 2511) and Qwen Image 2.1, plus experimental LTX-2 weights. Every release comes with its own pairing of base model, input order, prompt and workflow, so follow the guide for the version you load.

Who it suits: It suits image creators who use ComfyUI and want to swap a head or body using a reference photo. Use it only with faces and people who have agreed. The card asks you to keep out public figures and anyone who has not consented.

Industry | Who uses it | How they use it | Example
Creative | Image creators in ComfyUI | Replace a head with a reference face | A portrait that carries the reference person's head
Photography | Photo editors with consent from the people shown | Swap one body in a group photo | Replace only the person marked in magenta
Research | Teams testing image-editing LoRAs | Compare versions across base models | Run the Qwen Image 2.1 and Flux 2 Klein versions on one image

Anyone without consent from the people in the image, and anyone who needs guaranteed identical results, can skip it.

Where to use it:

• ComfyUI, with the workflow files the repository includes
• Base models: LTX-2, Krea 2, Flux 2 Klein, Qwen Image Edit and Qwen Image 2.1
• Windows, macOS or Linux, where you can download the files with Python

Use case | Who | What they get back
Head swap | Image creators | A picture with the head replaced using a reference face
Body swap | Photo editors | A picture with the body replaced
Selective swap | Photo editors | Only the person marked in magenta replaced, useful in group photos and busy scenes

How to set it up: You need these first:

• A ComfyUI install, kept up to date
• The base model, text encoder and VAE for the family you pick
• Python and the huggingface_hub package, to download the files

1. Pick a family from the card's list and read its guide in the repository.
2. Download the repository files with the Hugging Face tool.
3. Load the LoRA together with its matching base model and workflow in ComfyUI.

pip install -U huggingface_hub
hf download Alissonerdx/BFS-Best-Face-Swap

How to use it: Follow the input order in the guide for your version, since several head releases expect the body first and the reference face second. Start the LoRA strength at 1.0, then adjust it for your base model, resolution and reference image.

Tool | What it does
Qwen Image 2.1 LoRAs | Head swap, body swap and a selective body swap, with v1.1 as the recommended head file
Qwen Image Edit LoRAs | Face and head versions for the 2509 and 2511 editors
Flux 2 Klein LoRAs | Head swap versions for the 4b and 9b models
Krea 2 LoRAs | Head swap and body swap versions
LTX-2 LoRAs | Experimental head swap weights

Example: You load the recommended Qwen Image 2.1 head LoRA with its base model, put the images in the order the guide gives, and leave the strength at 1.0. You get the picture back with the head replaced by your reference. The card says results can differ from its demonstrations on other images.

For newcomers and experts: A LoRA is a small add-on file that changes what a large image model does. A good first step is to pick one family, read its guide, download the files and load the LoRA with the base model it was made for.

Each release carries its own contract for base model, input order, prompt and workflow. The Qwen Image Edit 2511 head files come in an original form and merged forms at rank 16 in fp16 and rank 32 in fp32. Flux 2 Klein has 4b and 9b versions, and the LTX-2 files are experimental. The card says to match each LoRA to its base model whenever possible.

Pick a base model family
Read that version's guide
Load the LoRA with its matching workflow
Set the input order and strength 1.0
Run it in ComfyUI

Good to know:

• Versions: for Qwen Image 2.1 the head file v1.1 is marked recommended, with an alternative and v1 also listed
• Results: the card says its examples only demonstrate the method, and identical output on every image is not promised
• Troubleshooting: update ComfyUI first, then make sure the LoRA, workflow, VAE, text encoder and base model all come from one family
• License: mit
• Cost to run: none named, no API key or account
• Data it reaches: it downloads files from Hugging Face and works on the images you give it in ComfyUI
• Not for: images of public figures, or of anyone who has not agreed to it, and the card asks you not to share such results

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Downloads in the last 30 days: 218,283
• Likes: 1,561
• License: mit
• Languages: en
• Based on: Qwen/Qwen-Image-2.1
• Tags: diffusers, lora, qwen-image, qwen-image-2.1, qwen-image-edit, face-swap, head-swap, body-swap, selective-body-swap, image-editing, comfyui, image-to-image

Links

• Model: https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap
• Files: https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Alissonerdx/BFS-Best-Face-Swap

Then follow the usage in its model card. Nothing in this kit runs by itself.