Base Model
The foundation a model or LoRA was built on — they must match, like LEGO pieces from the same set.
Every Model and LoRA is built on top of a base model — the foundation it was trained from. Base models matter for one big practical reason: a LoRA only works with models that share its base. Think of it like LEGO pieces — if they're from incompatible systems, they simply won't connect, and you'll get poor or broken results.
The families you'll see on PixAI
Illustrious family (SDXL) — Haruka, Hoshino, Otome series. Detailed anime character work.
Noob family (SDXL) — Hinata, Hikari. Vibrant colors and dynamic compositions.
Other SDXL bases — Pony, Animagine, and more.
SD 1.5 — the older generation. Not compatible with any SDXL LoRA.
DiT — Tsubaki and Serin. Use DiT LoRAs only. See DiT-Model.
How to stay compatible
Check the filter. In the Model-Market, filter LoRAs by base model (e.g. Illustrious) to match what you generate with.
Read titles and descriptions. A LoRA titled "[Illustrious] Komori Met" tells you its base right in the name.
Training your own? Pick the base model you actually generate with most — see LoRA-Training.
In the SD 1.5 era almost everything shared one base, so this rarely mattered. With SDXL and DiT, checking the base is the first debugging step whenever a LoRA "doesn't work."