♻ Loop (All Samplers)

"Which sampler should I use" is one of those questions everyone answers with a habit rather than a test. This node turns it back into a test. Wire it into your sampler_name input, set it to loop, and ComfyUI runs your whole graph once per sampler on the same prompt, same seed, same everything else - so instead of guessing, you get a side-by-side grid and can actually look.

How it works

Set sampler_name to ALL SAMPLERS and the node emits ComfyUI's full stock sampler list - around 44 of them, from euler and euler_ancestral through the DPM++ family, uni_pc, ddim, and more - as a single wildcard list. That triggers ComfyUI's native list execution: your KSampler and everything after it runs once per sampler, in order, with nothing else in the graph changing. Pick a specific sampler instead of ALL SAMPLERS and it just passes that one through, so the node doubles as a normal sampler-select widget when you don't want the full sweep.

Worth knowing before you queue a 44-way sweep: samplers aren't interchangeable in one dimension. Some (Euler, DPM++ 2M, UniPC) are converging - same seed gives the same settled image, and extra steps just refine it. Others (Euler ancestral, DPM++ SDE) are ancestral - they inject fresh noise each step, so they never fully settle and more steps changes the image rather than sharpening it. A same-seed sampler grid will show you real, meaningful differences, not just noise. And this node only cycles the samplers ComfyUI ships with - third-party solver packs like RES4LYF live outside this list entirely and won't show up no matter what you select.

The inputs and outputs that matter

There's exactly one input, sampler_name - either ALL SAMPLERS or one specific sampler from the stock list - and one output carrying whichever samplers got selected, feeding straight into your KSampler's sampler_name.

Pair it with the pack's sibling node, Loop All Schedulers, if you also want to vary the scheduler (Normal, Karras, sgm_uniform, and so on) - or use Loop Combos if you want the two swept together as matched pairs instead of every sampler crossed with every scheduler, which multiplies fast.

How to install it

ComfyUI Manager → search Bjornulf_custom_nodes → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/justUmen/Bjornulf_custom_nodes
pip install -r Bjornulf_custom_nodes/requirements.txt

Zero special dependencies for this node specifically - it's a static list plus ComfyUI's own sampler registry. The pack's heavier requirements (ffmpeg, ollama, whisper, and so on) exist for its other 150-odd nodes and don't affect whether this one loads.

Common issues & troubleshooting

44 renders is a lot of GPU time for one test. That's the real cost of this node, not a bug - every sampler in the list runs your full graph, and on serverless GPU billing that's 44 separate executions. If you're testing on comfy.icu, run a small resolution and low step count first to see which samplers are even in the right neighborhood, then re-run the shortlist at full settings.

Some samplers look identical, others look completely different. Expected, and it's the converging-vs-ancestral split above - don't assume a subtle difference on one pair means the whole sweep is subtle. Ancestral samplers especially will diverge more the more steps you give them, exactly backwards from what you might expect.

I want to compare against a third-party sampler I installed separately. This node only knows about ComfyUI's built-in list; it can't reach into another pack's registered samplers. You'd need to test those manually, one at a time, outside this loop.

Nothing looped - I got one image. Check that sampler_name is actually set to ALL SAMPLERS and not a specific pick; the node silently behaves like a normal single-sampler selector otherwise.