ADSchedulerSelector

ADSchedulerSelector is the pack's scheduler picker, and it is exactly as simple as it looks: a dropdown with twelve scheduler names, one AD_SCHEDULER wire out. You're choosing how the denoising steps march toward the final image - the noise schedule that decides when the model works on structure versus detail. It's the AsyncDiff pack's own version of the scheduler combo you already know from ComfyUI's KSampler, wrapped in the pack's wire type.

The twelve options are the classic diffusers menu: ddim, euler, euler_a, dpm_2, dpm_2_a, dpmpp_2m, dpmpp_2m_sde, dpmpp_sde, heun, lms, pndm, and unipc. Default is dpmpp_2m, and for once the default is a good one - it's the modern workhorse, good quality-to-step ratio, low noise, and it's what most workflows end up on anyway.

Which one should you pick?

Here's the honest translation for someone who doesn't live in sampling papers:

• dpmpp_2m (default) - start here. Fast, stable, high quality.
• dpmpp_2m_sde / euler_a / dpm_2_a - the "a" ones add stochastic noise. More variety in output, sometimes nicer detail, a bit less predictable. Good when results feel too samey.
• ddim / euler / pndm / lms - the older classics. Perfectly fine, just usually not better than dpmpp_2m for the same step budget.
• heun - the slow, picky one. Great quality, costs you steps. Only reach for it if you have time to burn or a specific look you're chasing.
• unipc / dpm_2 - the "somebody on the internet said this was better for X" zone. Try them, but don't let anyone convince you there's a magic scheduler that fixes a bad prompt.

Note that this is the pack's only scheduler control, and it only feeds two nodes: the scheduler input on ADSDSampler and ADSDUpscaleSampler. The video sampler (ADSVDSampler) and the AnimateDiff sampler (ADADSampler) don't take a scheduler input at all - if you're on a video workflow, this node simply isn't part of your graph. Don't go looking for the input that isn't there.

Install

The usual:

cd ComfyUI/custom_nodes
git clone https://github.com/SlackinJack/asyncdiff_comfyui.git
cd asyncdiff_comfyui
./clone_asyncdiff_repo.sh   # required - pulls the upstream AsyncDiff framework

Restart ComfyUI, or install via ComfyUI Manager by searching "asyncdiff_comfyui".

Where people get burned

Two things. First, this pack is multi-GPU by design - on a single card the sampler underneath will be slower than ComfyUI's native KSampler, and no scheduler choice fixes that. Second, keep your support expectations honest: the author's README states stable-video-diffusion-img2vid as the currently-supported path, and the SD text-to-image lane this scheduler serves is the less-tested side of the pack. The schedulers are all real diffusers schedulers doing real things; the docs just aren't the densest, so when in doubt, leave it on dpmpp_2m and move on with your life.