Burve Blind Grid Splitter

A "blind" grid splitter slices an image into rows × columns of tiles with zero content analysis. It doesn't look for faces, subjects, or seams - it just cuts on straight lines. If that sounds too simple to need a node, you're right, which is exactly why it's good at what it does.

The use case is chunking big images. Feed it a tall poster or a huge scan, get back a batch of smaller tiles you can push through whatever wants small inputs - an upscaler that chokes on large canvases, a detailer, or a model that works better on crops. The name "blind" is the feature: no subject detection means it's predictable and fast, but also means a subject that straddles a cut line will get split in half. Pair it with something that re-merges tiles, because this pack doesn't ship a joiner - you're on your own for reassembly.

How it works

Straight tensor slicing, no OpenCV. It takes the image's height and width, floor-divides by rows and cols to get tile dimensions, slices, and stacks every tile into a single batch - so a 2×2 split of one image comes back as four images in one IMAGE batch. No analysis, no dependencies beyond the pack itself.

The inputs that matter

• image - the source IMAGE. A batch in means every frame gets split, and all tiles from all frames end up in the same output batch, so keep an eye on that if you feed video frames.
• rows / cols - default 2, range 1–64. The grid is whatever you set; 64×64 on a small image will hand back nearly-1px tiles.
• center_crop - the only decision this node makes. If the image doesn't divide evenly, the leftover pixels get dropped. With center_crop off it drops from the bottom-right (top-left anchored); on, it centers the used area and trims both edges.

Output is one tiles output (an IMAGE batch), ready for a preview node, a batch splitter, or a per-tile processor.

Gotchas worth knowing

Because it uses floor division, a 1000px image split into 3 columns isn't 3×333 - it's 3×333 plus a dropped 1px strip. And center_crop doesn't rescale anything; it only decides which region gets used. One defensive detail in the code: if the grid would produce sub-1px tiles, the node just returns the original image instead of erroring, which can silently "do nothing" if you set a huge grid on a small image. If the tiles don't look like a full grid, check whether the image dimensions actually divide by your rows and cols.

Install via ComfyUI Manager or the standard git clone into custom_nodes, and restart. There's no per-node setup; it rides along with the rest of the Burve pack.