What is a Sampler?
A sampler is the algorithm that removes noise from a random field step-by-step until it becomes your image. Different samplers take different paths through "noise space," producing subtly different results even with the same prompt and seed.
Think of it like different routes to the same destination — some are faster, some are more scenic, some are more reliable.
The Main Samplers
Euler / Euler Ancestral — The classic. Fast, simple, good baseline. Euler Ancestral adds randomness at each step (more creative, less consistent). Good for: quick iterations, exploring ideas.
DPM++ 2M Karras — The current gold standard for quality. Produces clean, detailed images with good color accuracy. "Karras" uses a noise schedule that front-loads detail work. Good for: final renders, photorealistic content.
DDIM — Deterministic (same seed = exact same result every time). Faster than many samplers but sometimes softer output. Good for: reproducibility, animations where frame consistency matters.
UniPC — Unified predictor-corrector. Excellent quality in fewer steps (10-15 steps often enough). Good for: speed without sacrificing quality.
Quick Reference
| Sampler | Quality | Speed | Best For |
|---|---|---|---|
| DPM++ 2M Karras | Excellent | Medium | Final renders |
| Euler | Good | Fast | Quick tests |
| Euler Ancestral | Good (varied) | Fast | Creative exploration |
| DDIM | Good | Fast | Consistency |
| UniPC | Very Good | Fast | Low-step generation |
Our Recommendation
Start with DPM++ 2M Karras at 25 steps. This is the default on EGAKU AI and produces consistently excellent results across all model types. Only switch if you have a specific reason to.