I trained a flux LORA and it works great, but it has a really hard time generating images of the lower half of my body. I assume this is because I trained it mostly on close-up/mid shots. Has anyone experienced this issue before?
1 Like
Hey, any progress or success on this? I am trying to train LoRA for full body portrait.
1 Like
I had the same issue, until I got repetitive with the prompt. That, and I combined LoRA’s of me, I trained them differently, one was of just my name (on it’s own, the clothing follows you on every prompt which is annoying), but it got my face right every time. The other was describing everything about the image, except me.. what I mean is, describe hair, describe clothing, scene, lighting, distance from camera, etc, but not my ethnicity, not my facial features, Things that don’t change on you, eye color for example, you don’t train it on, via caption. Then I load both LoRA’s in at 0.50 & 0.50, or sometimes 0.45 & 0.45, and be sure to specify in more ways than just 1, "Full-body shot of a man/woman/robot standing with their feet firmly planted on the ground, with the feet spread apart at least body width, and the full height from head-to-toe visible in a high definition 9:16 (or whichever) ratio taken in vertical (or horizontal, but that usually cuts off unless you say “taken at a distance” or “viewed from afar”) format.
then, when the prompt is ending, add an additional reinforcement to it, “Showing the full length of the subject, a full-body photograph to capture the subject in their entirety”.
1 Like