Type | |
Stats | 34 11 |
Reviews | (5) |
Published | Sep 19, 2024 |
Base Model | |
Training | Steps: 550 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | lawren |
Hash | AutoV2 59DCB30B01 |
Experimental LoRA to capture the distinctive style of Lawren Harris' paintings of the Canadian North. The best way I can describe it is an abstracted (i.e. not fully abstract) impasto (i.e. not fully smooth), depth/contour-"slice" perspective style.
Perhaps because of this style, it is an unusual LoRA/dataset in that the first training epoch (Epoch 1) gave the most acceptable result and further training epochs were way off.
On photo-realistic images, it does not recreate the painting style but instead rearranges the composition.
LoRA Weights
When prompting for images similar to the paintings' subject matter (e.g. mountains, glaciers etc.) the optimal weight range tends to lie in the middle around 0.4-0.6.
However for other subject matter, higher weights are generally better depending on your choice of Sampler.
