April 14, 2026 · 3 min read

Local Flux undress guide (Flux Kontext + Undress LORA) – Low end GPUs supported

TL;DR

This is a setup guide for a ComfyUI workflow that uses FLUX Kontext models to remove clothing in images. It tells you which model files to download, where to place them in the ComfyUI folders, which version to choose based on your GPU VRAM, and which workflow JSON to import. Low-end GPUs are told to use quantized GGUF files and an extra LoRA/plugin. The basic flow is: install ComfyUI, download the required encoders/VAE/model files, put them in the right directories, install ComfyUI-GGUF if needed, load the matching workflow, upload an image, adjust the prompt, and generate.

This guide assumes some knowledge of ComfyUI – use this tutorial if you need more help setting up ComfyUI: https://comfyui-wiki.com/en/install/install-comfyui/comfyui-desktop-installation-guide.

Required Downloads

models/enconders: https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/clip_l.safetensors
models/vae: https://huggingface.co/Comfy-Org/Lu…aged/blob/main/split_files/vae/ae.safetensors
models/loras: https://pixeldrain.com/u/VdYgEpN6

Download depending on your GPU

For 8GB VRAM
models/unet: https://huggingface.co/bullerwins/FLUX.1-Kontext-dev-GGUF/blob/main/flux1-kontext-dev-Q4_K_M.gguf
models/text_encoders: https://huggingface.co/comfyanonymo…blob/main/t5xxl_fp8_e4m3fn_scaled.safetensors

For less than 8GB VRAM or older generation cards like the 1080Ti
models/unet: https://huggingface.co/bullerwins/FLUX.1-Kontext-dev-GGUF/blob/main/flux1-kontext-dev-Q4_K_M.gguf
models/text_encoders: https://huggingface.co/comfyanonymo…blob/main/t5xxl_fp8_e4m3fn_scaled.safetensors
models/loras: https://huggingface.co/alimama-crea…blob/main/diffusion_pytorch_model.safetensors

For over 16 VRAM
models/diffusion_models: https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/blob/main/flux1-kontext-dev.safetensors
models/text_encoders: https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/t5xxl_fp16.safetensors

Note for low end gpu owners: if you’re seeing long generation times or OOM errors, you need a more quantized version (smaller file size) from https://huggingface.co/bullerwins/FLUX.1-Kontext-dev-GGUF/tree/main

You will also need ComfyUI-GGUF. You can install this through Comfy Manager, or simply install all the missing modules from workflow.

This is what your file structure should look like:

📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── flux1-kontext-dev.safetensors
│   ├── 📂 unet/
│   │   └── flux1-kontext-dev-Q4_K_M.gguf (if dogshit GPU)
│   ├── 📂 vae/
│   │   └── ae.safetensors
│   ├── 📂 loras/
│   │   ├── flux_turbo_alpha.safetensors (if really dogshit GPU)
│   │   └── JD3s_Nudify_Kontext.safetensors
│   └── 📂 text_encoders/
│       ├── clip_l.safetensors
│       └── t5xxl_fp16.safetensors (or t5xxl_fp8_e4m3fn_scaled.safetensors if dogshit GPU)

Download the Required Workflow:

comfy_ui_nudify_workflow.json – https://pixeldrain.com/u/HH38tVpR
comfy_ui_nudify_workflow_gguf.json – https://pixeldrain.com/u/HAoDnTee
comfy_ui_nudify_workflow_gguf_turbo.json – https://pixeldrain.com/u/DMLAXi2f

Conclusion

Upload your image, tweak the prompt, and you’re good to go. This works far better when you’re more detailed: for example, if the subject is wearing a white top then you should specify white top.

Example generated using comfy_ui_nudify_workflow_gguf_turbo.json with the caption underneath: