GR Sigma Preset Selector

SIGMAS is the noise schedule a sampler consumes - how aggressively noise gets removed at each step, kept separate from the sampler algorithm itself in ComfyUI's design. Reshaping that curve is normally the kind of thing power-user packs like RES4LYF exist for, with a whole menu of solvers and schedules behind it. This node does a friendlier, narrower version of the same idea: pick a preset that matches your subject, tell it how many steps you're actually running, and get back a ready-to-use sigma curve plus a graph so you can see what you're about to commit GPU time to.

How it works

preset is the headline input, and unlike its sibling GR Sigma Presets - which offers ten fixed img2img/motion-strength curves with no step control at all - this one has 26 choices organized around subject matter and intensity: zimage (the default), balanced, composition_heavy, detail_heavy, aggressive, subtle, portrait, landscape, architecture, abstract, fine_detail, fast_decay, and more beyond that. zimage being the default is a real tell - GraftingRayman ships other Z-Image-specific tooling too, so it's a fair bet these presets were tuned with that model in mind first.

steps tells the node how many sampling steps you're actually running, and auto_scale_to_steps (on by default) reshapes the chosen preset to fit that count rather than using whatever step count the preset was originally authored at. Leave it on unless you specifically want the preset's native length regardless of your actual step count - turning it off means a mismatch between steps and what you get out will look wrong, since the preset won't have been rescaled to match.

The inputs and outputs that matter

• preset - which of the 26 subject-matched curves to use.
• steps - how many steps you're sampling at.
• auto_scale_to_steps - rescale the preset to fit steps; leave this on unless you know why you'd turn it off.

Three outputs: sigmas (SIGMAS, wire to your sampler), graph_preview (IMAGE, a plotted view of the actual curve - your sanity check before a real run), and preset_info (STRING, a readback of what you picked - useful for a text-display node or just confirming what ran).

How to install it

ComfyUI Manager, search GraftingRayman. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/GraftingRayman/ComfyUI_GraftingRayman

Restart afterward. The whole pack fails to import without OpenAI's CLIP installed separately, even though this node never touches CLIP - portable ComfyUI: .\python_embeded\python.exe -m pip install git+https://github.com/openai/CLIP.git; system Python: pip install git+https://github.com/openai/CLIP.git.

Common issues & troubleshooting

Pack won't load. Check your startup console for the CLIP import error and run the pip install above.

Result looks off - too much or too little structure preserved. Preview graph_preview before you trust a preset name blindly; subject-matched names are a starting point, not a guarantee, and the graph tells you exactly what shape you're actually running.

Curve doesn't seem to match your step count. Confirm auto_scale_to_steps is on. With it off, the preset's native length wins regardless of what steps says.

Need to fine-tune pacing beyond the preset itself, not just the subject match. That's what GR Sigma Preset Selector Advanced adds - the same 26 presets, plus a curve_shape override. For full manual control over the curve's zones, go to GR Sigmas.