Denoise Slider

Denoise is the single most important number in img2img, and this node exists so the Flux+ workflow has exactly one place to set it. In ComfyUI Flux Continuum - Rob's all-in-one modular workflow - every operation that starts from an existing image routes through this slider: inpainting, outpainting, the ultimate upscaler, and the detailer all read their denoise from here. Mess this up and nothing else matters, so it's worth knowing what the dial is actually doing.

How it works

Denoise strength controls how far the sampler walks away from the source latent. At 0.0 the sampler starts from your image as-is - no change, useful for checking your setup. At 1.0 it starts from pure noise, which is effectively text-to-image wearing an img2img costume. Everything between is a blend.

The classic bands, straight from the inpainting playbook: 0.0–0.3 is subtle correction (color fixes, small cleanups), 0.3–0.6 is where most real edits live (swap an outfit, change an expression), 0.6–0.8 makes big structural changes, and 0.8+ regenerates the region almost entirely. The most common beginner mistake is cranking denoise to 0.7+ for a small detail fix and wondering why the whole masked area came back different.

Inputs and outputs

The node is a single slider:

• value - FLOAT, 0.0 to 1.0, default 0.5, step 0.001. That's the whole input.

Output is one FLOAT carrying the number straight through - the node is a pass-through with a nicer UI, not a processor. In the workflow that float feeds whatever sampler chain is active, so flip the Output selector to inpainting or upscaler and this same value drives it. Nothing to rewire per module, which is the entire point of the pack's unified control interface.

Installing it

This ships inside the robertvoy/ComfyUI-Flux-Continuum repo, so install the pack, not the node. ComfyUI Manager → search "Flux Continuum", or:

cd ComfyUI/custom_nodes
git clone https://github.com/robertvoy/ComfyUI-Flux-Continuum

Restart ComfyUI after. The pack has zero Python dependencies of its own, but the full Flux+ workflow wants Impact Pack, rgthree, ComfyUI Essentials and KJNodes, plus BFL's dev models (fill, depth, canny, flux1-dev) and the ae.safetensors VAE. Bypass the module loaders you don't use and you can skip those downloads.

Troubleshooting

Two things trip people up. First, the slider's range (0–1) means it only works as an img2img-style denoise - if you wire it into a plain txt2img KSampler, you're feeding a KSampler a denoise value that does nothing, because txt2img doesn't use denoise at all. Second, on Flux specifically, low denoise with a high step count wastes most of the steps refining pixels the sampler already believes are final. If your inpaint looks "samey" no matter the slider, try raising denoise before touching steps. And if your image looks overcooked, remember the guidance side of the workflow auto-bumps to 30 for inpaint/canny/depth - that's the pack's design, not a bug.