SDXL has hundreds of checkpoints. Picking the right base model matters more than stacking LoRAs. This article gives the mainstream picks for four major scenarios.
Four scenario picks
| Scenario | Mainstream checkpoints | Notes |
|---|---|---|
| Realistic portrait | RealVisXL, JuggernautXL, ZavyChromaXL | Strong skin texture and detail |
| Anime | Animagine XL, Pony Diffusion XL, ArthemyComics XL | Anime-specific training |
| Landscape / cinematic | DreamShaper XL, NightVision XL, CineNostalgia XL | Atmospheric, strong color grading |
| Illustration / stylized | SDXL base + style LoRAs | Few stylized base models; LoRA-led |
Selection criteria
- Check author update frequency — actively maintained models stay stable
- Check license — CreativeML Open RAIL++M generally permits commercial use
- Check sample gallery — confirm the look matches your expectation
- Check LoRA compatibility — some bases conflict with specific LoRAs
Common stacking strategy
Professional workflow:
- 1 base model (chosen direction)
- 1–3 LoRAs (character + style + detail)
- Optional 1 textual inversion for specific looks