
Version 2.0 update:
after I connected the grid pattern sometimes visible in bright or dark parts of the images with the fp4 training with OneTrainer, I switched to fp8 training, which got rid of the pattern. But I have to redo all my LoRAs to fix them all. The dataset included two different poses: one where the individual is lifting the upper body up by leaning on the elbow (not the outstretched hand, that is Leaning to Side ). And another where the person is lying flat on the ground, oftentimes resting the head on the hand. It wasn't possible to combine these two into one LoRA, so I split them in two, this is the original, propped up version. So, the keywords are lying on side and leaning on elbow . As always, a LoRA tag like <lora:LyingOnSide-step00600:1> may help.
Additional note: I spent a lot of time making dozens of test images to decide which epoch works best and provides the best images. After publishing I tried to made a portrait format image with a new prompt and the 20 epoch version failed totally. Even if I don't want to do it, I have to pile on the 25 epoch version which was very close. Decide yourself.
Flux is notoriously bad at creating anything which is lying down. The only pose I got was lying on the stomach and then the feet are wrong. I tried first to do a all-in-one solution, but after a lot of testing I had to hammer each pose into a separate LoRA.
I also noticed that you have to use a <lora:LyingOnSide-150:1> tag, otherwise it is more or less chance if it will influence your image. The trained sentence is "lying on her side". Here the two variations are actually in one LoRA, but with less control: more propped op or lower, with the face resting on the hand. You can try "her head ist resting on her hand" or "is leaning on her elbow" , that was my captioning scheme.
Something I started here but had to be placed too in separate LoRAs to work is a finer control over face angles, body angles, camera angles and image area. But starting with this one, the captions are consistent as a preparation, even if they do not (yet) work standalone. The scheme is:
Face angles:
Full face view | Three-quarter face view | Two-third face view | Profile face view
Body angles:
Front angle shot | Three-quarter angle shot | Side angle shot | Three-quarter rear angle shot
Camera angle:
High camera angle | Eye level shot | Low camera angle | Shoulder level shot | Cowboy shot | Ground level shot
Image area:
Full-body portrait | Three-quarter-body portrait | Half-body portrait | Quarter-body portrait
It would be great if I get this to work!
描述:
Version 2.0 is trained with fp8, getting rid of the grid pattern sometimes visible in the images. The dataset was pruned to only include the images with the upper body lifted up.
训练词语: lying on side,leaning on elbow
名称: LyingOnSide-step00600.safetensors
大小 (KB): 37531
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success