Gigachad SamplerCustomAdvanced

Gigachad SamplerCustomAdvanced is the pack's power-user sampler, and unlike the main KSampler it doesn't hide behind a sampler dropdown - it expects you to bring the pieces. noise, guider, sampler, and sigmas are all inputs you construct with other nodes, which is exactly how you build the chained and custom-guidance sampling graphs that a plain KSampler can't express. Think of it as the stock SamplerCustomAdvanced with four extra conveniences bolted on.

The four add-ons

• noise_multiplier (default 1) - scales the noise tensor before sampling. Scale it up and you push a pass toward more variation; 1.0 means unchanged. The code wraps the noise generator so the scaling applies to whatever noise node feeds it.
• cfg_rescale (default 0, off) - the tooltip says it plainly: "CFG rescale strength (0 = off). Reduces over-saturation on Flux/SD3." If you've ever fought blown-out colors on a high-CFG Flux or SD3 run, this is a targeted fix that doesn't require a separate rescale node.
• preview_method - none, auto, latent2rgb, or taesd. Live step previews built in, so you don't need to drop a preview node into the middle of your sampler chain.
• return_with_leftover_noise - keeps the noise level at whatever step you stopped, so the output can feed directly into a second sampler pass. This is the key to proper chained sampling.

The standard advanced controls

Beyond those, it mirrors the familiar SamplerCustomAdvanced inputs: start_at_step (skip the first N sigmas), end_at_step (stop early; 10000 = run all), the latent_image, and the outputs output and denoised_output. Wire the denoised_output of a first pass into the latent_image of a second with return_with_leftover_noise on, and you've got a two-stage refinement without any bookkeeping nodes.

When you'd actually use this

Three concrete cases. First, multi-pass sampling: low-step first pass for composition, then a second pass with a different guider or sigma range for detail. Second, custom guiders: if you're using a guide node like the Flux/SD3 rescale guiders, this is where they plug in. Third, Flux/SD3 over-saturation: cfg_rescale is a one-knob fix that the stock advanced sampler makes you wire a patch node for.

If you never leave the land of plain KSampler + a prompt encode, this node is overkill - skip it. If you've been accumulating "how do I chain samplers" bookmarks, this is a solid home base.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/comfyui-gigachad.git

or install comfyui-gigachad via ComfyUI Manager and restart. No requirements.txt - pure ComfyUI, no extra packages. The only prerequisite is that you already know your way around guider/noise/sigmas nodes, because that knowledge is what makes this node useful rather than confusing.