Trying to mimic flat offset print in 1960s (primary colors + flat colors). Experimenting merging a number of Loras with trials and errors.
First time doing merging loras. Prefer not to, but took painfully long for training Flux1D one.
it took 5 hours per epoch on my 3060Ti. 100 img x 10 repeats x 6 epochs + NetDim 32 took 3 days. Gave up.
The metadata in images shows the original Loras used to merge into this one. Instead of having all of them, only this one will do.
Set Strength: 0.8
PS. The tool used for merging Loras https://github.com/Anashel-RPG/anashel-utils.git
Description
FAQ
Comments (1)
This LoRA is way too large, but it works decently well if you turn down the strength. At 1.0 it tends to get a bit messy and the images lose some coherency. Also recommend a high guidance of at least 3.5, even up to 4.5