This LoRA was trained on 57 vibrant and colourful "happy" oil pantings, so it is quite good at this type of images.
So be happy and have fun with it! 😁🎈
Tip: if you want something more "painterly", try lowering the Guidance from the default 3.5 to 1.5 or 2.0.
Description
This is epoch 11, other epochs can be downloaded from tensor.art/models/853190650487689837
FLUX.1 - dev-fp8
Trigger: painting
Repeat: 20 Epoch: 3 (Trained on 57 512x512 image for 3420 total steps)
Then for another 6 epochs for total steps (* 20 9 57)=10260
Finally, training for another 2 epochs but reduced to 49 images
(took out mostly landscape images) (* 20 2 49)=1960
So total number of steps is (+ 1960 10260)=12220
Unet LR: 0.0005 LR Scheduler: cosine Optimizer: AdamW
Network Dim: 8 Alpha: 4
Epoch Loss
1 0.436
2 0.422
3 0.413
4 0.414
5 0.398
6 0.388
7 0.402
8 0.387
9 0.369
10 0.362
11 0.349<-lowest