Trying for a better version of the re-flat loraCarry on, nothing to see

Still flat

Still_flat_r1.safetensors

I absolutely wasn't happy of my last try, so... Another one, yeah.One version trained for NOOB (this time for real, on Mol-Keun deep blue), theoretically softer. Sadly I fear there are art bumps in the output :-/The other one finally has what I was searching for, I needed higher learning. Sadly it seems there is some kink (the bad ones, I mean)

Reflat soft

Latest version trained with NAI chenkinProbably the best one. A bit weak, but quite stable: in the test I cranked the weight up to 2.5 without bad consequences, so go for it

Files with SHA256: b8e556e6f5d8d58753a298ca5f44789996019b081d9b2e7fb049c9ec7e1e04eb