FLUX LoRA for generating a female subject side by side - dressed on the left side and completely nude on the right side. Trained on female subjects only.
How to Use: No trigger words for this one. Trained with images that have verbose, natural language captions so works best with similar style prompting. Prompts should contain general information about the subject (age, ethnicity, body type, hairstyle, etc), detailed description of clothing (for the left side) and detailed description of body for the right side. You can check sample images for the prompt templates.
Sampling Steps: 25 - 30
Sampling method: Euler or Heun
Schedule type: Simple or Beta
DCFG: 2.5 - 3.0
LoRA weight: 1.0
Slight differences between the two sides (pose, facial expression, hair, lighting) are not necessarily a failure of the LoRA. The training data was full of these little differences between the left and right sides. So it's natural. As long as it looks like the same person the LoRA has done it's job. With that said - could take a bit of trial and error to get almost perfect results, but that's to be expected.
Description
FAQ
Comments (1)
There are many of these LORAs, but interestingly no successful one that solves the 'front', 'side' and 'back' view - Flux-Dev is capable of both (without LORA, but obviously not with full nudity), BUT the intentional prompt nerfing means that most times, the three pose request fails (one pose is repeated).