Bilibin was an illustrator artist. The part of his style is a frame with ornament, but in practical cases I don't need it. Training the Lora with frames could bring a lot of troubles, so I cut them off in the dataset.

For common use, set strength '1.0', negative '3d, render, photo, cgi', and 'source_cartoon' (recommended but not always necessary). For using with pony checkpoint's baked character or with Lora, increase the strength at least to '1.2', same negative and 'source_cartoon' are highly recommended, otherwise the 3D effect of a character will 'overpower' the Lora in most cases. Adetailer/only masked inpainting is required. For adetailer I recommend setting the detection value as low as possible, like in mostly 2D pictures — adetailer's detection is not really good.