SamplerDPMPP_2M_SDE_SUN
The most popular SDE sampler, with uniform noise — and the solver_type question
- SAMPLER
If you ever ran a ComfyUI workflow in 2023–2024 and picked a sampler at all, there's a good chance it was DPM++ 2M SDE - the multi-step stochastic solver that became the community's "just works" choice on SD 1.5 and SDXL before DPM++ 2M (non-SDE) took over for pure speed. SamplerDPMPP_2M_SDE_SUN was that sampler with one alteration: the noise injected along the SDE path was scaled uniform noise on [-sigma_down, +sigma_down] instead of the usual Gaussian, seeded deterministically from your sampling seed.
And, like every other node on this family of pages: it no longer exists. The July 2024 "Complete rework, breaking changes" removed all six per-sampler nodes from bvhari/ComfyUI_SUNoise. Today the pack only ships the selector nodes SamplersSUNoise and SamplersSUNoiseAdvanced, both offering dpmpp_2m_sde.
The input the others didn't have
This node had three inputs: eta (1.0), s_noise (1.0), and solver_type - midpoint or heun. The 2M SDE solver uses a second-order correction term from the previous denoised state, and solver_type picks how that correction is computed. Heun is the default and the smoother of the two; midpoint is marginally cheaper and marginally rougher. In practice, most people couldn't tell them apart, which is exactly why the plain SamplersSUNoise selector just hardcodes heun and only the Advanced selector lets you touch it.
Output was a SAMPLER for SamplerCustomAdvanced.
Fixing a workflow that references it
ComfyUI will list it as a missing node. Replace it with SamplersSUNoise, set sampler_name = dpmpp_2m_sde, choose noise_type. Want solver_type back? Use SamplersSUNoiseAdvanced instead - it exposes all four knobs. The current code also interleaves SUNoise with standard Gaussian every other step, which the 2024-era version of this node didn't do; that's the v2.0.0 improvement and it's a genuine reason not to pin the old commit unless you have to.
Install, for the record
cd ComfyUI/custom_nodes
git clone https://github.com/bvhari/ComfyUI_SUNoise
Restart, or use ComfyUI Manager → search "SUNoise". No extra dependencies, no model downloads, GPL-3.0. The pack gets essentially zero community discussion, so there are no forum threads of known issues - the version trap is the entire troubleshooting section.
Bottom line
You're on this page because a workflow from the pack's first quarter references this class. The fix is a two-click swap to SamplersSUNoise → dpmpp_2m_sde, and the modern version is strictly better (interleaved noise). Keep the pack's standing guidance in mind: 30 steps or more for good quality, and if you're comparing samplers, fix the seed - an SDE sampler never settles, so wandering results are the feature, not a bug.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| solver_type | COMBO | 2 options: midpoint, heun | |
| eta | FLOAT | 1.000–100 | — |
| s_noise | FLOAT | 1.000–100 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SAMPLER | SAMPLER | — |