Sampling Methods - Best Use & Best Settings

Sampling Methods - Best Use & Best Settings _ DPM++ 2M Karras Best for: General use, LoRAs, stable anatomy, clean output Steps: 20–30 Scale: 7–9 _ DPM++ SDE Karras Best for: Painterly, softer, organic results Steps: 30–40 (37 is a valid effective value) Scale: 6–8 _ Euler Ancestral (Euler a) Best for: Stylized, anime, creative variation Steps: 15–25 Scale: 6–8 _ Euler Best for: Fast, deterministic, clean results Steps: 20–30 Scale: 7–9 _ Euler Dy Best for: Dynamic detail emphasis Steps: 20–30 Scale: 7–9 _ Euler SMEA Dy Best for: Enhanced edge/detail handling Steps: 25–35 Scale: 7–9 _ Euler Negative Best for: Experimental / niche workflows Steps: 20–30 Scale: 7–9 _ Euler Negative Dy Best for: Experimental dynamic negative guidance Steps: 25–35 Scale: 7–9 _ DDIM Best for: Consistency, reproducibility Steps: 20–25 Scale: 7–9 _ Heun Best for: Smooth gradients, conservative refinement Steps: 30–40 Scale: 6–8 _ KDPM2 Ancestral (KDPM2 a) Best for: Creative variation, sharper transitions Steps: 20–30 Scale: 6–8 _ KDPM2 Best for: Deterministic, cleaner results Steps: 25–35 Scale: 7–9 _ LMS Best for: Older workflows, smooth convergence Steps: 25–40 Scale: 7–9 _ LMS Karras Best for: Improved LMS stability Steps: 25–40 Scale: 7–9 _ DPM2 Karras Best for: Stable classic DPM behavior Steps: 25–40 Scale: 7–9 _ DPM2 a Karras Best for: Creative, more stochastic DPM2 Steps: 20–30 Scale: 6–8 _ DPM++ 2S a Best for: Fast stochastic DPM++ generation Steps: 15–25 Scale: 6–8 _ DPM++ 2S a Karras Best for: Cleaner stochastic output than 2S a Steps: 20–30 Scale: 6–8 _ DPM++ 2M Best for: Deterministic DPM++ without Karras Steps: 20–30 Scale: 7–9 _ DPM++ SDE Best for: Organic refinement without Karras schedule Steps: 30–40 Scale: 6–8 _ DPM++ 2M SDE Best for: Balanced stochastic refinement Steps: 30–40 Scale: 6–8 _ DPM++ 2M SDE Exponential Best for: Aggressive early refinement Steps: 30–40 Scale: 6–8 _ DPM++ 2M SDE Karras Best for: High-quality stochastic refinement Steps: 30–40 Scale: 6–8 _ DPM++ 2M SDE Heun Best for: Very smooth, conservative refinement Steps: 35–45 Scale: 6–8 _ DPM++ 2M SDE Heun Karras Best for: Maximum stability + smoothness Steps: 35–45 Scale: 6–8 _ DPM++ 2M SDE Heun Exponential Best for: Heavy refinement with controlled noise Steps: 35–45 Scale: 6–8 _ DPM++ 3M SDE Best for: Very gradual, high-fidelity refinement Steps: 40–60 Scale: 6–8 _ DPM++ 3M SDE Karras Best for: Maximum quality at high Steps Steps: 40–60 Scale: 6–8 _ DPM++ 3M SDE Exponential Best for: Strong early convergence at high detail Steps: 40–60 Scale: 6–8 _ DPM fast Best for: Speed over quality Steps: 10–20 Scale: 7–9 _ DPM adaptive Best for: Automatic Step control Steps: Auto Scale: 7–9 _ Restart Best for: Experimental workflows, multi-restart denoising Steps: 66 (baseline recommended by the sampler’s creator) Scale: 7–9 Important: Restart is designed for very high Steps. Using low Steps breaks its intended behavior. _ PLMS Best for: Legacy workflows Steps: 20–30 Scale: 7–9 _ UniPC Best for: Modern, efficient, high-quality output Steps: 20–30 Scale: 7–9 _ KohakuLoNyuYog Best for: Anime-focused Kohaku workflows Steps: 20–30 Scale: 7–9