Trained on Civit AI. The dataset of ≈650 images was carefully compiled by me and captioned using my end-to-end captioning workflow, which basically extracts the original prompt, uses JoyCaption to write new prompts, and then deploys GPT as a refiner to synthesize the two into a final caption (I've been meaning to write an article about this because I think some others would find it useful). All images in the dataset were AI-generated, featuring various artistic illustrated styles but sharing a ligne claire comic book vibe.