Trained for ~75 steps off a single image.

Training caption was:

colored skin, multiple girls, profile, red sclera, blue skin, red eyes, portrait, yellow eyes, looking at another, eyelashes, 2girls, lips, orange eyes, colored sclera

So using any of those will trigger it to some degree. Colored skin has the most pronounced effect. I highly recommend including locations, poses, etc in your prompt or else it will compose your image similar to the training image. The 'Colored skin' tag can be applied to anything, it will work on objects, scenery, etc. The sample images I've provided have been tested at a variety of weights. Going past 2 yields poor results.

I'm actually quite pleased with this and I think I'll dig around and try this with other images I've generated in the past.