Sampler Euler Ancestral Fixed Noise 🍌

Euler Ancestral (Euler A) is the sampler people reach for when they want variety rather than a perfectly reproducible result - it's an ancestral sampler, meaning it injects fresh random noise at every single step instead of settling toward one answer. That's exactly why it looks livelier than plain Euler, and exactly why it's a pain to debug: because the per-step noise itself isn't reproducible, changing one unrelated setting can shift the whole image even on the same seed, which breaks the standard "change one thing, keep everything else fixed" comparison workflow. This node is cgem156's fix for that specific annoyance - Euler A's character, with the noise pinned down.

What's different from stock Euler A

The tell is the noise parameter, a combo defaulting to "fixed." Where the schema only documents that one option, the implication is clear from the name: it locks the per-step ancestral noise draws to something reproducible tied to your seed, instead of the usual fresh-random-every-step behavior stock Euler A uses. You keep the ancestral look - the visible variety that makes people choose Euler A over a converging sampler in the first place - but two runs with the same seed and settings should actually match now, which they don't with the vanilla version.

eta and s_noise are the standard ancestral-sampler knobs underneath: eta controls how much stochastic noise gets injected at each step (0 pushes the sampler toward behaving like plain, non-ancestral Euler; higher values add more variation from the base trajectory), and s_noise scales the magnitude of whatever noise does get added. Both default to 1, which is the normal Euler A behavior - this node's value is entirely in the noise mode switch, not in reinventing the math.

Using it

Output is a SAMPLER - wire it into KSampler (Advanced)'s sampler input, or SamplerCustomAdvanced if you're building the pipeline from the pack's other pieces (this repo also ships TCDSampler and TextScheduler, which slot into the same kind of manual pipeline).

Installing it

Via ComfyUI Manager: search cgem156-ComfyUI, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/laksjdjf/cgem156-ComfyUI

Restart ComfyUI afterward. No model downloads, no extra dependencies - it's a pure sampler implementation.

Common issues & troubleshooting

Results still aren't reproducible run to run. Double-check noise is actually set to fixed - that's the entire point of this node over stock Euler A, so if it's set to anything else you've lost the fix.

Output looks like plain Euler, not ancestral Euler. Check eta isn't sitting at or near 0 - that dials the ancestral noise injection down toward nothing, which flattens the "livelier" look Euler A is known for.

You're trying to A/B test a LoRA strength or a prompt tweak and results are still moving unpredictably. Reproducible noise only fixes the sampler's own randomness - it doesn't make the rest of the pipeline deterministic. If your comparison is still noisy, check whether something else upstream (a random-seed noise generator, a non-fixed CFG guider) is the actual source of the variance.

You don't actually want reproducibility. If you liked stock Euler A's behavior and just found this node by accident - it's fine, this is strictly an add-on for the debugging/comparison use case. If you're not trying to A/B test, ordinary Euler A does the same creative sampling without the extra combo widget.