← Help Center
Fast Photorealistic (Recommended)
Best for: photorealistic images, artistic compositions, general purpose generation
- FLUX.2-klein-4B — High quality, fast inference (4 steps, CFG 1). Best balance of speed and quality
- FLUX.2-klein-base-4B — Base variant of FLUX.2 klein, also optimized for speed
Recommended settings: Steps 4, CFG 1.0, Euler a
LUMINA Models
Best for: photorealistic images with more detail, artistic photography
- Neta-Lumina — Fast photorealistic generation with good detail
- NetaYume-Lumina-Image-2.0-v40 — Enhanced Lumina variant with improved quality
Recommended settings: Steps 40, CFG 4.0, Euler a
Anime & Illustration (SDXL)
Best for: anime characters, manga-style art, illustrations, fantasy art
- SDXL-animij_v5 — Popular anime illustration model
- SDXL-autismmixSDXL_autismmixPony — Versatile anime generation
- SDXL-ArliMix-v1 — Curated mix with excellent quality
- SDXL-illustriousXL_smoothftSOLID — Smooth, high-quality anime output
- SDXL-zukiBestAnimeMix_v10 — Best anime mix from the zuki series
- SDXL-sudachiXLIllustrious_v1 — Illustrious-based with wide style range
Recommended settings: Steps 20, CFG 6.0, Euler a
Vision-Language Models
Best for: image understanding, description, editing with language guidance
- Z-Image — General purpose vision-language model (steps 30, CFG 4)
- Z-Image-Turbo — Fast variant for quick results (steps 8, CFG 1)
- Qwen-Image — Qwen vision model for image tasks (steps 30, CFG 1)
Image to Image & Inpainting
Any model works for img2img. The quality depends on the model and denoising strength:
- Low denoising (0.1-0.3): Best for minor edits, color changes, style adjustments
- Medium denoising (0.4-0.6): Good for significant transformations while keeping composition
- High denoising (0.7-0.9): Major overhaul, new composition inspired by the original
Sampler Guide
| Sampler | Best For |
| Euler a | General purpose, good default for all models |
| DPM++ 2M Karras | High quality, works well with fewer steps |
| DPM++ SDE Karras | Artistic/creative outputs, slower but more varied |
| DPM++ 2M SDE Karras | Highest quality, best for detailed prompts |
| DDIM | Stable, predictable results |
| UniPC | Fast convergence, good for low step counts |
Auto-Settings
The playground automatically applies each model's recommended sampler, step count, and CFG scale when you select it. You can always override these values manually.