Auto Grid / Stitch (one image)

The simplest node in the GridSplit pack, and proud of it. Give it one image and it tiles that same image into every cell of an R×C grid - 1×2, 2×2, all the way up to 4×4 - then scales the whole result to a target total resolution. That's the whole job. No cropping, no distortion, no hidden model download, nothing to tune if you don't want to.

When would you actually reach for it? A few honest reasons: you want a repeating grid or contact-sheet look from a single image, you're testing how a composition holds up when tiled (grid wallpaper, texture preview), or you're building a workflow that needs a grid to start from one source before other nodes operate on the whole grid at once. Because it runs the exact same stitch engine as the pack's advanced multi-image node, it's also a convenient way to sanity-check that engine with a boring input before you wire up sixteen real images.

How it works

Each cell gets a copy of your one image. The stitching replicates ComfyUI's built-in "Stitch Images" behavior in its match_image_size mode: within a row, every copy is resized to match its neighbor's height with aspect preserved and no padding, rows are normalized to a shared width, and the result is a single clean rectangle. Only after that does it scale to your target megapixels - and note this is the total size of the whole grid, not per cell. Default is 4 MP, which for a 2×2 grid leaves each cell around 1024×1024-equivalent; raise it if the tiles look soft.

The rows and columns are set with the little 4×4 picker drawn inside the node - hover to preview a block, click to lock. Cells are numbered top-left origin, row-major, so a 2×2 is 1/2 on top and 3/4 below. The picker writes the same rows and cols values you could also type by hand; it's just the friendly way to set them.

Inputs and outputs

• image - the one image to repeat into every cell.
• rows, cols - grid dimensions, 1–4, via the in-node picker.
• megapixels - target total grid size (0.1–100, default 4).
• scale_method - resample filter, lanczos by default. Leave it alone.

Outputs are grid (the stitched IMAGE), plus width and height - the final pixel dimensions, handy if something downstream needs to know the exact size without recalculating it.

Install

Same story as the rest of the pack: no model files, no extra pip packages, just torch and numpy which ComfyUI already ships with. ComfyUI Manager → search "GridSplit", or:

cd ComfyUI/custom_nodes
git clone https://github.com/workordie/ComfyUI-GridSplit

then restart ComfyUI and hard-refresh the browser (Ctrl+Shift+R on desktop). That last step matters more than usual here: the in-node grid picker is a small frontend widget served from the pack's web/ folder, and a cached page won't show it. If you see the node with no picker, that's the reload you skipped - one of the most common "this node is broken" reports for this pack, and it's never the node's fault.

If you think you'll ever stitch more than one image, just grab the Advanced sibling instead - it does everything this one does and then some. This is the quick path for the single-image case.