KSamplerOverrided

KSamplerOverrided is the working sampler that pairs with KSamplerSetting. One node holds your baseline settings as a DICT; this one actually runs the sampling pass, and you only wire in the fields you want to change for this particular pass.

Think of it like a config file with command-line overrides. Everything defined in the setting dict applies by default, and each optional input on KSamplerOverrided - seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise, and even model - replaces its counterpart when you connect a wire. Leave a socket empty and the dict's value is used as-is.

That's the whole trick, and it's genuinely handy. A hires-fix setup becomes "share one settings dict, override steps and denoise on the second pass" instead of mirroring values across two samplers. It also makes a tidy pattern for trying variations: keep the dict, swap one override, re-run.

Mechanically it's a thin wrapper over ComfyUI's common_ksampler - same sampling pipeline, same latent output (LATENT) you'd feed into a VAE Decode. Feed its output to VAEDecodeBatched and SaveImage exactly like any other sampler.

Install with ComfyUI Manager (search "ComfyUI-nodes-hnmr") or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/CYBERLOOM-INC/ComfyUI-nodes-hnmr

Then restart ComfyUI. No model downloads, no extra Python deps. This is the maintenance fork of hnmr293's pack, kept alive because the original stopped merging fixes and broke with newer ComfyUI.

The one trap: because every input is optional, it's easy to build a graph where you think you're overriding something and the wire isn't connected, so the dict value quietly wins. When a "changed" setting has no effect, check your wire first. Also note the setting dict carries latent_image and conditioning, so if you wire a different positive/negative here, make sure the downstream conditioning actually connects - this node takes the override, but it doesn't merge prompts for you.

Reach for this whenever you're running the same sampler twice with small differences. If you never need per-pass overrides, a plain KSampler is simpler. If you want to sweep many values at once rather than override one, that's what KSamplerXYZ is for.