NAI XY 轴参数

You want to know which guidance value looks best, or whether a sampler change matters, or how a prompt suffix changes the vibe. The pack's XY system answers that with a grid: change one parameter across N values, keep everything else fixed, and lay the results side by side. NAIXYAxisV2 is the node that declares a single axis of that grid - which parameter, and which values. Two of these feed the plot node, one for X and one for Y.

What it is

Three inputs:

• parameter - the thing you're varying. The list is long and covers the settings that actually matter: Seed, 采样步数 (steps), Guidance, CFG Rescale, CFG Delay, 采样器 (sampler), 噪声调度 (noise schedule), 模型 (model), 宽度/高度 (width/height), 正面提示词后缀/负面提示词后缀 (prompt suffix), and - the ones people forget - 全部Vibe强度, 全部Vibe信息提取量, 全部精准参考强度, 全部精准参考忠实度 (sweep every attached Vibe/precise-reference's strength or fidelity at once).
• values - the values to try, one per line (default 4.0\n5.0\n6.0). A JSON array also works. Between 1 and 6 values - the plot node enforces a maximum 6×6 grid.
• axis_label - optional; if empty it just uses the parameter name as the label.

Outputs: the XY轴 object (NAI_XY_AXIS, feed this into the plot node), 值数量 (INT - how many values it parsed), and 轴摘要 (STRING - a human-readable version of the axis).

How to think about it

This node does zero work itself - no API call, no Anlas. It's a data structure: "on this axis, for each of these values, override parameter P." The plot node reads it and builds the grid. Two of these per plot, X and Y. If you only want a one-dimensional sweep, the pack lets you set one axis to a single value, but honestly the interesting comparisons are the 2D ones - guidance × sampler, model × steps, seed × prompt suffix.

The 全部Vibe... and 全部精准参考... parameters are the sleeper hits. Instead of manually re-running a workflow with Vibe strength at 0.3/0.5/0.7, you declare that as an axis and get all three side by side with identical everything else. That's the kind of comparison you'd otherwise swear you'd get around to and never do.

Install and setup

Part of ComfyUI-NovelAI-GENYTOOLS. Manager search, or:

cd ComfyUI/custom_nodes
git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git
pip install -r ComfyUI-NovelAI-GENYTOOLS/requirements.txt

Restart, set your NovelAI Persistent API Token under NovelAI 2.0: Persistent API Token.

Gotchas

• More than 6 values is a hard error at the plot stage, not a warning. Keep axes ≤ 6 or the plot node refuses.
• Values are parsed leniently (lines, commas, or JSON), but keep them clean - a stray 5.0, can parse as a string and silently misbehave on numeric parameters.
• The 值数量 output is your sanity check: wire it to a display and confirm the axis parsed the way you intended before you let the plot node spend money.
• The axis node is free. The plot node downstream is where each cell becomes a paid generation.