You can find the ComfyUI workflow for properly using this LoRA model in the model gallery.

Trigger prompt: fchlpr, Side-by-side amateur photos: a close-up portrait and a half-body shot of a woman

The gallery was generated using NF4 + hyperflux LoRA for 8 steps.

More examples can be found in the gallery under the model.

This experimental LoRA model works but is far from perfect, as facial similarity is not always achieved.

This approach is quite resource-intensive, requiring 12GB+ for generations using the workflow from the gallery. Lower resolutions will result in poorer image quality. I am unsure about how to proceed with high-resolution (Hires) generation afterward.

The training dataset and configurations are attached to the model.