This LoRA is intended for use when editing images with Flux 2 Klein models.

This LoRA guides Flux 2 Klein models to generate images that more closely match the subject in the given input image. In particular it's trained on image pairs where the subject changes pose and face angle. This model was trained exclusively on images of people, but it appears to work decently well with other characters based on the limited testing I've done.

Examples were generated using FLUX 2 Klein Base 9B, with 50 steps, 1 CFG, 1 Megapixel, LoRA Strength 1.6

• Higher steps often work better

• Some seeds work better than others

• Recommended Strength for FLUX 2 Klein Base 9B is 1.6

• Recommended Strength for other FLUX 2 Klein models I've tested is 2.4