⚡| Image Grid Builder

Image Grid Builder answers a specific, universal itch: you've generated eight seeds, six styles, or a whole batch of LoRA tests, and now you need them on one canvas to actually compare them. This node takes an image list and lays it out into a grid - with an optional text banner and full control over spacing, columns, and colors. It's the contact sheet node that lives in the Z-Image Power Nodes pack, but nothing about it is Z-Image-specific; if you've ever screenshot five ComfyUI previews and stitched them in paint, this replaces that ritual.

How it works

The input is an autogrow image list - the node starts with two image slots and adds more as you connect them (up to 50). Every connected image gets resized by image_scale (default 0.5) and laid out left-to-right, wrapping at columns (default 0 = unlimited, meaning one long horizontal row). The optional banner mode (none/header/footer) reserves a strip at the top or bottom and renders your text into it, with its own height, font, size, and colors. Everything comes back as a single IMAGE.

The inputs that matter

• images - the autogrow list. This is the whole point: connect your batch output (or several individual previews) here.
• columns (INT, 0–64) - 0 means one long horizontal row. Set 2–4 for a proper comparison grid.
• banner (COMBO) + text (STRING) - header/footer with your label. This is what makes a seed test readable later, so use it.
• image_scale (FLOAT, default 0.5) - shrinks each cell; 1.0 keeps them full size (and blows up your memory fast).
• image_spacing (INT, default 8) - gap between cells, in pixels.
• banner_height (default 144), font_name, font_size, text_color, lines_color, banner_color - the banner styling. Hex strings, e.g. #FFFFFF.

Output is one IMAGE - wire it to a preview or Save Image.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/martin-rizzo/ComfyUI-ZImagePowerNodes

Restart ComfyUI, or install through ComfyUI Manager (search "Z-Image Power Nodes"). No pip dependencies. Note the pack's minimums: Python 3.10, PyTorch 2.5, ComfyUI v0.11.0.

Where it pays off

Z-Image Turbo has a famous quirk: different seeds produce near-identical images, so the interesting comparisons are style vs style and LoRA strength vs LoRA strength, not seed vs seed. A grid with a banner naming each column is the fastest way to eyeball those. Practical tips: keep image_scale at 0.5 or below for anything more than four cells, and set columns explicitly - the unlimited default turns an eight-image batch into a comically wide strip. One honest caveat: it arranges, it doesn't annotate per-cell, so the banner is your only text - name your test before you run it.