Detail Daemon Graph Sigmas

Detail Daemon Sampler has a stack of knobs - bias, exponent, the two offsets, fade, the curvature stuff - and unless you can picture what they do, you're basically tuning blind. Graph Sigmas is the little companion node that draws the picture. It plots the adjusted sigma curve right on the node so you can see how your settings reshape the noise schedule instead of guessing and re-rendering. That's the whole job. It's a debugging and learning tool, not part of your final graph.

What it does and, importantly, doesn't do

Feed it a sigma schedule and the same Detail Daemon parameters, hit queue, and it renders a graph of the resulting curve on the node face. It had to be a separate node from the sampler to work at all - the sampler can't both adjust sigmas and draw them mid-run. The one thing to burn into memory: this node does not change your generation. It only visualizes. Nothing it does touches the image; it's a read-out. Set the values here to match your Detail Daemon Sampler and you get a preview of what that sampler is doing to the schedule, which is genuinely useful the first few times you're learning how start, end, bias and exponent interact.

Because this is pure visualization with no rendering cost, it's a cheap way to build intuition - worth wiring up once while you learn the pack, then deleting from the graph.

The inputs

It mirrors Detail Daemon Sampler almost exactly, so the ones that matter are the same: detail_amount, start, and end, with bias, exponent, start_offset, end_offset, fade and smooth shaping the curve. Set them identical to your actual sampler or the graph is telling you about a run you're not doing.

Two differences from the sampler node are worth flagging. First, instead of a sampler input it takes a sigmas input of type SIGMAS - you feed it a real noise schedule, typically the output of a scheduler node like BasicScheduler, so the graph reflects your actual step count. The README's tip is to drive both this node and the sampler from the same primitive nodes so their values can't drift apart. Second, where the sampler has cfg_scale_override, this node just has cfg_scale (default 1).

This is an output/display node, so it has no output connections - nothing wires out of it. You run the queue and look at the node.

Installing it

It's part of the Detail Daemon pack, so it arrives with the others. Via ComfyUI Manager, search ComfyUI-Detail-Daemon and install; or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Jonseed/ComfyUI-Detail-Daemon

Restart ComfyUI. No models, no dependencies.

Common snags

The most common confusion is expecting it to do something to your image. It won't, ever - if you added it hoping for more detail, you want Detail Daemon Sampler instead; this one just draws graphs. Second: you have to actually run the queue to see the curve. It doesn't update live as you drag values; queue a run and the graph redraws. Third, if the graph looks nothing like what your sampler is producing, your inputs have drifted out of sync - the fix is to feed both nodes from the same source values rather than typing them twice and hoping they match.