Sampler and Scheduler Compatibility Matrix

Choosing the right sampler and scheduler combo is kind of like picking the right shoes for a marathon — you can wear flip-flops, but don’t act surprised when you trip at Step 12. Below is your cheat sheet for pairings that actually perform well — no guesswork, no flaming garbage results.

Best Sampler + Scheduler Compatibility Matrix (Quick View)

Sampler	normal	karras	exponential	sgm_uniform	simple	ddim_uniform	beta	linear_quadratic	kl_optimal
euler	✅
euler_cfg		✅
euler_ancestral			✅
euler_ancestral_cfg_pp
heun
heunpp2		✅
dpm_fast
dpm_adaptive
dpmpp_2s_ancestral
dpmpp_2s_ancestral_cfg_pp
dpmpp_sde
dpmpp_sde_gpu
dpmpp_2m		✅
dpmpp_2m_cfg_pp							✅
dpmpp_2m_sde		✅
dpmpp_2m_sde_gpu
dpmpp_3m_sde								✅
dpmpp_3m_sde_gpu
ddpm
lcm				✅
ipndm						✅
ipndm_v
deis					✅
res_multistep		✅
res_multistep_ancestral
re_multistep_ancestral_cfg_pp
gradient_estimation
gradient_estimation_cfg_pp							✅
er_sde			✅
seeds_2
seeds_3
ddim
uni_pc									✅

✅ Quick Legend:

• ✅ = Best known scheduler pairing for this sampler.
• Blank = Not recommended / niche use / no clear benefit pairing.

---

Pairing the right scheduler with the right sampler in ComfyUI isn't just a “nice to have” — it's the difference between buttery-smooth masterpieces and noisy, incoherent messes. While most samplers technically work with most schedulers, that doesn’t mean they should. Each sampler has unique mathematical characteristics — some prioritize precision, others speed, others realism — and the scheduler determines how that sampling process unfolds over time. The wrong combination can undermine your output quality, tank performance, or worse, make your beautifully engineered workflow behave like it just rolled out of a chaos factory. Choosing the best pairings ensures you get faster generations, better detail retention, smoother gradients, and more consistent results — especially in high-stakes workflows like SDXL, animations, or multimodal conditioning. Trust us: aligning your scheduler with the sampler’s strengths is the easiest quality boost you can make without touching a single prompt.

📚 Detailed Best Pairing List

Sampler	Best Scheduler	Why This Pairing Works	Sampler Docs	Scheduler Docs
euler	normal	Fast and sharp results, good for sketch-style or high-contrast work.	euler	normal
euler_cfg	karras	Maintains CFG-weighted detail well, stable under long prompts.	euler_cfg	karras
euler_ancestral	exponential	Best for dreamy, soft lighting and slow transitions.	euler_ancestral	exponential
dpmpp_2m	karras	High-quality, well-balanced — the industry gold standard.	dpmpp_2m	karras
dpmpp_2m_cfg_pp	beta	CFG-enhanced DPM++ with excellent edge preservation.	dpmpp_2m_cfg_pp	beta
dpmpp_2m_sde	karras	Fantastic for realism; handles shading and depth extremely well.	dpmpp_2m_sde	karras
dpmpp_3m_sde	linear_quadratic	Complex scene generation, rich gradients, great for SDXL.	dpmpp_3m_sde	linear_quadratic
heunpp2	karras	Cleaner transitions between token weight shifts, good for intricate prompt detail.	heunpp2	karras
lcm	sgm_uniform	Optimal fast sampler; pairs with low step configs.	lcm	sgm_uniform
uni_pc	kl_optimal	Adaptive and smart. Excels at high-resolution and SDXL workflows.	uni_pc	kl_optimal
deis	simple	Very clean, progressive sampling. Pairs well with text-to-image.	deis	simple
ipndm	ddim_uniform	Great compromise for noise-controlled diffusion steps.	ipndm	ddim_uniform
res_multistep	karras	Works well for animations and sequential inference.	res_multistep	karras
gradient_estimation_cfg_pp	beta	Smooth transitions, precise edge definition for CFG-heavy workflows.	gradient_estimation_cfg_pp	beta
er_sde	exponential	Best used for SDXL variants and 3D-looking renders.	er_sde	exponential

🧩 Notes on Exclusions

• ddpm, seeds_2, seeds_3, dpm_adaptive, and dpm_fast were excluded for being legacy/utility samplers or having no strong "best" pairing — they work, but aren't ideal for quality-first workflows.
• If you don’t see a combo listed here, assume it’s okay but not optimal unless you have a very specific reason to use it.
• We’re skipping raw CFG samplers unless you're explicitly building a custom pipeline that depends on parallel prompt/latent conditioning.

🧯 What-Not-To-Do-Unless-You-Want-a-Fire

• ❌ Pair lcm with exponential, kl_optimal, or linear_quadratic. It's meant for speed and doesn't behave well with over-complicated schedulers.
• ❌ Use uni_pc with simple or ddim_uniform unless you like flat, lifeless outputs.
• ❌ Stack CFG samplers (*_cfg_pp) without a prompt setup that supports dual CLIP encoders. You'll lose all that enhanced guidance precision you paid for.
• ❌ Apply dpmpp_sde_gpu with high noise schedulers (exponential, ddim_uniform) unless you're tuning for chaos.

📚 Additional Resources

• 🔗Scheduler Options
• 🔗Sampler_Name Options