Testing Flux.1 Lora training. Used my Dataset for my SD1.5 Nox Lora using the setting Citiva recommends. Just a test on it.

EDIT: my computer has quite a Hard time running even the lower versions of Flux.1 I cant get many Pictures due to running FLux + a Lora jumps from my GPU into my shared Memory. 15 minutes per Image isnt exactly nice... Now i need a new GPU...

Edit:Edit: After a few updates of Forge memory Usage is far better no longer getting my GPU VRAM maxxed out. The model seems to work as intended with some Caveats such as the size of the crystal and needing to play with the Lora weights more. Flux has Major Potential We need Fine Tunes to pull it out. also high CFG is odd. Only Increment to 1.5 MAX Otherwise CFG should just be left at 1.