Z-Image Turbo FaceSwap - ComfyUI & One Click Windows Installer

If you've been wanting an easy way to swap character faces using your favorite LoRAs, this workflow makes it effortless. With the Z-Image Turbo and Meta's SAM 3 models powering everything, you can take any image and blend characters together using flexible LoRA adjustments for style and accuracy.
The idea builds on the original img2img workflow, but now it's focused specifically on head-swapping for character edits. You'll mainly be adjusting three things: denoise level for how much it changes, CFG scale for stronger image guidance, and LoRA strength to match how your model was trained. Everything's built right into the workflow, so setup is quick.
Original creator post on Reddit: Reddit | Original workflow: Pastebin
 
Preloaded Models (Low VRAM)
z_image_turbo-Q5_K_M.gguf model — Hugging Face
Qwen3-4B-UD-Q5_K_XL.gguf text encoder model — Hugging Face
z_image_turbo_vae.safetensors (ae.safetensors) VAE model — Hugging Face
2xLexicaRRDBNet_Sharp.pth upscale model — Hugging Face
Note: The standard Z-Image Turbo Diffusion Models (BF16) are not bundled with the installer, but can be downloaded from the Comfy Org Hugging Face repository. ComfyUI Manager: GitHub.
 
System Requirements
Nvidia RTX 30XX/40XX/50XX GPUs (FP16 supported)
CUDA-compatible GPU (minimum 4 GB VRAM, 8 GB+ recommended)
Windows OS
At least 40 GB free storage
A Z-Image Turbo Character LoRA
 
What's Included
Portable ComfyUI Windows Installer, pre-configured for Z-Image Turbo Image-to-Image SAM detection Character Face Swapping
Automated downloads for all models and nodes
User-friendly workflows for both newcomers and advanced users
Fast Groups Bypasser support for quick workflow toggles
 
Usage Notes
Load the provided workflow in ComfyUI.
Ensure all checkpoints are correctly assigned in your Loaders, especially your Character LoRA.
Upload the target image to the Load Image node.
Add any extra prompting in the Add important extra info here node near the bottom.
Adjust resolution and aspect ratio carefully to avoid VRAM overloads above 1024 x 1024.
Video tutorial for creating a character LoRA: https://youtu.be/uHqOdGsR9cw