Player not loading? Watch on YouTube
This tutorial builds a ComfyUI workflow that uses a reference face to generate new images of a similar-looking character. The speaker describes it as image generation rather than a face swap and demonstrates IP-Adapter FaceID PlusV2 with a Juggernaut XL checkpoint.
Setup requires CLIP ViT-H, a FaceID PlusV2 model, and its matching LoRA for SDXL or SD 1.5. The supplied notes specify renaming the CLIP download to CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors. The files go into ComfyUI's clip_vision, ipadapter, and loras model folders. The tutorial installs ComfyUI_IPAdapter_plus through ComfyUI-Manager, then installs insightface and onnxruntime in the same virtual environment that runs ComfyUI.
The workflow connects IPAdapter Unified Loader FaceID and IPAdapter FaceID to a loaded reference image. A preview node shows the extracted face, and the SDXL example uses a 1024 by 1024 latent image. The loader applies the matching LoRA automatically.
The speaker adjusts FaceID strength and selects strong style transfer, reporting better hair reproduction but some artifacts under the eyes. They also report better results when the reference image comes from the same checkpoint. The demonstrated setup targets SDXL and SD 1.5; the speaker cautions that Pony and Illustrious models may behave differently.
