Pony & F1.Dev LoRAs to generate in the style of traditional Chinese paintings!
The only necessary trigger word is "shukezouma", but you can try adding "traditional media" or "calligraphy".
It's been given images of portraits of people, sceneries, landscapes, plants, and animals. Some of these images are AI-generated (selected for best quality or adherence to style), but the majority come from original media.
The Pony version works best with 1024 x 1024, but I am not sure if the FLUX version shares this preference. The models have been given data from a wide array of aspect ratios, so it shouldn't matter too much, although it may generate unnaturally large trees if the aspect ratio is too large (i.e trees the size of a mountain).
Using scores in Pony may decrease style adherence when prompting for faces (it tries to make the subject realistic or in anime/3D), but otherwise works well for prompting landscapes and animals.
Otherwise, using high scores may also remove the "textured" canvas appearance typically seen in these paintings.
Finally, typical upscaling techniques also diminish the style adherence of the final image (it tries to clean up the brush strokes or the small dotted details), so it is best to upscale it in img2img to maintain the style and brush stroke detail.
You may need to use the Civitai generator or otherwise specific webUIs that support FLUX LoRAs (e.g. ComfyUI). The provided images for the FLUX version are the sample prompts generated during training.
Recommended weight: 1