Miyo Chinatsu from Crimson Comics.

So, what can I say—this character was an experiment for me, since Miyo appears only in a black-and-white doujinshi. For her, I decided to try creating a minimal dataset using images taken directly from the doujinshi, and then using the color cover—cut into multiple pieces—to make up at least half of the images. This way, the model can also generate color images.

My best advice for generating in color is to put monochrome and greyscale in the negative prompt.

I know her face isn’t perfectly accurate here either, but I think this time it’s mostly because the dataset mixes both colored and non-colored images.

One last thing: with the tags brown belt and gold buckle, you can also control the gold pendant on her neck. If you want to remove it, simply leave those tags out.

Now I’ll leave the rest to you—what do you think of the result?