SamplerDPMPP2SAncestral_SUN
The fan-favorite DPM++ 2S ancestral, now with uniform noise — and a warning
- SAMPLER
SamplerDPMPP2SAncestral_SUN was the SUNoise version of DPM++ 2S ancestral - the second-order ancestral sampler that displaced plain DPM2 and became an A1111-era favorite for its combination of speed, stability, and the creative wiggle ancestral samplers give you. The "warning" part: it's gone. This node was one of six per-sampler nodes in bvhari/ComfyUI_SUNoise that the July 2024 "Complete rework" deleted, replaced by the SamplersSUNoise / SamplersSUNoiseAdvanced selector nodes. Current installs of the pack (v2.0.0) don't ship it.
What it was
DPM++ 2S ancestral does a half-step prediction, then a full update from the second-order result - the "2S" is single second-order step per iteration, which is why it's cheaper than its multi-step siblings while still beating Euler a's accuracy. This node kept all of that and swapped the injected Gaussian noise for scaled uniform noise bounded to [-sigma_down, +sigma_down], deterministically seeded from the sampling seed. Same SUNoise rationale as the rest of the pack: Gaussian's extreme tail values are blamed for high-CFG oversaturation, and a bounded distribution avoids them.
Inputs were eta (default 1.0) and s_noise (default 1.0), output a SAMPLER for SamplerCustomAdvanced.
Why this one deserved better luck
This was arguably the most sensible candidate in the whole pack for the SUNoise treatment. DPM++ 2S a is a light, fast sampler that people already reach for when they want ancestral variety without DPM++ SDE's slowness - and the thing that made people stop reaching for ancestral samplers is precisely the noise-induced speckle and drift. Uniform noise was the right medicine for that symptom. The rework didn't kill the idea, though: the selector node still has dpmpp_2s_ancestral in the dropdown, with both noise_type options, plus the v2.0.0 interleaving that alternates SUNoise with standard Gaussian so you don't lose all per-pixel character.
Fixing an old workflow
If ComfyUI flags this node as missing (it'll be in any workflow saved between April and July 2024), swap it for SamplersSUNoise with sampler_name = dpmpp_2s_ancestral. Or pin the pack to efc6e38 for the exact old behavior - there's no good reason to. The selector version is the same sampler on current code, and it's the version that gets interleaved noise.
Install (for the record)
cd ComfyUI/custom_nodes
git clone https://github.com/bvhari/ComfyUI_SUNoise
Restart. Or use ComfyUI Manager → search "SUNoise". No dependencies, no model downloads, GPL-3.0.
Bottom line
You're here because an old workflow or old tutorial pointed at this node. The modern answer is SamplersSUNoise with dpmpp_2s_ancestral selected, and the honest modern take is unchanged from 2024: this is a nice, cheap way to get ancestral variety with slightly cleaner noise. Run 30+ steps as the README advises, and fix your seed if you're A/B-ing - ancestral samplers never settle.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| eta | FLOAT | 1.000–100 | — |
| s_noise | FLOAT | 1.000–100 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SAMPLER | SAMPLER | — |