Sampler Advanced v1

A sampler wrapper that mostly exists to leave you alone

Here's a rare honest pitch: this node does not invent a new sampler. Under the hood it delegates the actual diffusion loop to ComfyUI's SamplerCustomAdvanced - same noise, same guider, same sigmas, same sampling core. What it adds are two knobs for long video runs, and for the pack's own SuperNodes it's the canonical sampler stage. If you've built LTX-2 video graphs, you know why that matters: a ten-minute denoise shouldn't be fighting the frontend over progress-bar updates.

The two knobs

• disable_progress (default on) - turns off the per-step progress/UI update spam. The author's claim is that long video denoises feel smoother, and it's plausible: every step callback in a big LTX-2 latent is work the browser has to chew on. You lose the pretty bar; you keep the result. If you want to watch a long run's progress, flip it off.
• cleanup (default off) - after sampling, runs ComfyUI's soft empty cache plus torch.cuda.empty_cache(). Genuinely useful on low-VRAM rigs where the latent's peak allocation leaves the next stage (VAE decode) gasping. It costs a little time, so it's opt-in.

The inputs and outputs

The required inputs mirror SamplerCustomAdvanced exactly - noise, guider, sampler, sigmas, latent_image - so it drops into any graph that's already wired for advanced sampling: scheduler → sampler select → guider → this node. Outputs are output (the sampled latent) and denoised_output (the fully-denoised latent, the one you feed to VAE decode when you want the last clean step rather than the final noisy-ish output). It also copies the latent before sampling rather than mutating upstream objects, which saves you a class of "my graph changed after the sampler ran" surprises.

When to reach for it

Honestly, most people will encounter this node inside an IAMCCS SuperNode workflow rather than searching for it - it's listed in the SuperNodes requirements as a required helper, and it's the sampler those wrappers call. Standalone, it's a mild quality-of-life upgrade over stock SamplerCustomAdvanced: on a 500-frame LTX run, disable_progress genuinely reduces jank, and cleanup has rescued more than one 8GB card from an OOM on the decode step. If neither knob appeals, the stock node is fine and you lose nothing by staying.

Install is the pack install - ComfyUI Manager, search "IAMCCS", or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Then restart. No model downloads, no pip extras. The one real trap: it won't fix your sampling quality. If your video looks bad, this node is not the place to tune - same core, same results, just quieter. Pair it with a proper guider/sigma setup and treat cleanup as your VRAM safety valve.