A LoCon model for creating a more painterly, impressionist style for the base SDXL 0.9 model. In some cases, it may be better to not run the refiner model on the outputs, as it changes the art style a bit. Gonna update this thing when SDXL 1.0 releases, or, obviously, when I do more training. Have fun!
For best results, I recommend to start your prompt with "a painterly illustration of..." or "a digital painting of...". This guides the model to create a less detailed, softer appearance.
This model is designed for use with SDXL, and will not work with SD 1.5, 2.1, or any other version.
Description
Apparently Kohya trainer doesn't support building a model from weights for LyCORIS? This version is a re-training based on normal LoCon, trained on the same dataset for 50 epochs, and then further fine-tuned on just the term "digital art" for another 20. This model produces similar results, and may be consistent than the original one. Also, 300 megs -> 18!