ComfyUI Node

Image Tile

Chop a giant image into overlapping tiles, and reassemble them

By wenchengxiang·Created about a month ago·Updated a day ago· 1
Image Tile
  • 图像
  • 切片批次
  • 切片宽度
  • 切片高度
  • 重叠横向
  • 重叠纵向
行数2
列数2
重叠比例0.00
重叠像素_横向0
重叠像素_纵向0

There's a hard wall in ComfyUI: your GPU can only sample at so many megapixels, and a 4K image simply doesn't fit through a normal sampler. The classic way around it is tiling - split the image into pieces, process each piece on its own, stitch them back. Image Tile is the splitting half of that workflow, and it's one of the genuinely useful nodes in this pack.

It takes an image and chops it into 行数 (rows) × 列数 (columns) tiles, with an optional overlap. Overlap is the secret ingredient - when you tile without it, every seam is a visible line where one tile's lighting met another's. With overlap, adjacent tiles share pixels, and a good stitch node can feather the shared region so seams vanish. Here you can set overlap two ways: as a ratio (重叠比例, 0–0.5, e.g. 0.25 = a quarter of each tile) or as explicit pixel counts per axis (重叠像素_横向 / 重叠像素_纵向). The pixel version is capped at half a tile, and both collapse to zero when a dimension is 1×.

The outputs are designed for a hand-off. You get the tile batch plus three metadata values: 切片宽度 (tile width), 切片高度 (tile height), and the two overlap numbers. That's precisely the information needed to rebuild the image, and the pack's Image Untile node is waiting for it - feed it the tile batch plus rows, columns, and the overlap values and it reconstructs the original with feathered seams. It's the same crop-and-stitch logic the whole ecosystem runs on, packaged for big images instead of small regions. Also note the labels are in Chinese (行数, 列数, 重叠比例…), a quirk of this pack you'll just get used to.

Inputs and outputs that matter

  • 图像 (in) - the image to slice.
  • 行数 / 列数 (in) - tiles per axis, default 2×2.
  • 重叠比例 (in) - overlap as a fraction of tile size, 0–0.5.
  • 重叠像素_横向 / 重叠像素_纵向 (in) - overlap as explicit pixels.
  • 切片批次 (out) - the tile batch. 切片宽度 / 切片高度 / 重叠横向 / 重叠纵向 (out) - metadata for Image Untile.

Installing it

Part of wenchengxiang/ComfyUI-Practical-Tools. ComfyUI Manager → Custom Nodes Manager → search "ComfyUI-Practical-Tools" → Install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/wenchengxiang/ComfyUI-Practical-Tools

then restart. No pip dependencies, no models - pure Python.

Issues and gotchas

The classic beginner trap is overlap of exactly 0 - the tile workflow works, but every stitch line shows. Give yourself 0.1–0.25 overlap and let Image Untile's feathering do its job. Also know that the tile count multiplies fast: 4×4 is 16 tiles, and if you're running each through a heavy process that's 16 passes. And the metadata outputs are not optional decoration - if you lose them, you can't reconstruct without re-deriving the layout by hand. Keep tile, width, height, and overlaps wired straight into Image Untile and the round trip stays honest.

CategoryPractical-Tools/Image

Inputs (6)

NameTypeDefaultDescription
图像IMAGE
行数INT21–256
列数INT21–256
重叠比例FLOAT0.000–0.5
重叠像素_横向INT00–8192
重叠像素_纵向INT00–8192

Outputs (5)

NameTypeDescription
切片批次IMAGE
切片宽度INT
切片高度INT
重叠横向INT
重叠纵向INT