Illustrious

Illustrious (Illustrious-XL) is an open-source, SDXL-based text-to-image model developed by OnomaAI Research and optimized specifically for illustration and animation. Rather than training from scratch, it was built on the Kohaku XL-Beta (Revision 5) checkpoint, inheriting a strong anime-oriented SDXL foundation and then refining it for cleaner linework and tag adherence. Because it keeps the SDXL architecture and CLIP-based text encoding, it reads comma-separated Danbooru-style tags rather than the long natural-language prompts that transformer models like Flux expect. The technical details are published in the team’s arXiv paper (2409.19946).

The model is deliberately positioned as an open-collaboration project: OnomaAI released the weights openly and asks the community to keep improvements in the open rather than folding them into closed, proprietary products. That openness is a big reason the Illustrious lineage became a foundation others build on — most notably NoobAI-XL, which starts from Illustrious and extends its training on a broader Danbooru/e621 dataset. On Civitai the base Illustrious-XL checkpoint is hosted and ready to run, so you can generate from a tag prompt without downloading the ~6.5GB of weights.

Choose Illustrious when you want clean, tag-driven anime and illustration with strong character consistency and precise control over pose, outfit, and composition through booru tags. It anchors one of the largest anime LoRA ecosystems anywhere, so styles and characters are easy to layer on. If you want the same booru-native approach with even deeper character coverage, its descendant NoobAI is worth comparing; if you want photorealism, legible in-image text, or long natural-language prompts, a Flux-class model is the better fit. For general-purpose versatility outside anime, the parent SDXL ecosystem covers more ground.