ComfyUI Extension: ComfyUI Pad to Multiple

Authored by mr-september

Created

Updated

0 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Pads images to multiples of N (right/bottom only) with solid color fill. Useful for VAE/model dimension requirements.

Looking for a different extension?

Custom Nodes (0)

    README

    ComfyUI Pad to Multiple

    A minimal ComfyUI custom node that adds one node: ImagePadToMultiple.

    What it does

    Pads an image by default on the Bottom and Right so both width and height are divisible by a number you pick (default 16). You can independently select up to 2 directions (Top, Bottom, Left, Right, Left-Right, Top-Bottom) to pad. Uses a solid color for the padding. Nothing gets resized, cropped, or stretched.

    Most diffusion and transformer models need input dimensions divisible by specific values, like 64 for SD 1.5 / DiT, 8 for SDXL, 16 or 32 for Flux / SD3. If your image is 513×513 it won't work as-is, and the common workarounds (stretch, crop, resize-then-crop) all silently destroy your image.

    Padding is the right answer. The colored border shows you exactly where the padding is, so you can crop it off later in Photoshop, GIMP, or whatever you use.

    Why not just resize or crop?

    They sound fine until they aren't:

    • Resize (stretch): Warps everything. Faces, text, architecture all distorted.
    • Center crop: Cuts off the edges. Your subject might be near the border. Gone.
    • Resize + crop: Combines both problems. First warps, then chops.

    The padding approach keeps your original content untouched. The padded pixels are a solid color you chose, so you can find them and remove them trivially.

    Node: ImagePadToMultiple

    Category: image/transform

    Inputs

    | Input | Type | Default | What it does | |---|---|---|---| | image | IMAGE | — | The image to pad | | multiple_of | INT | 16 | Divisor target (1–256). Set to 64 for SD 1.5, 8 for SDXL, etc. | | pad_color | enum | "black" | Padding color: black, white, or gray | | direction_1 | enum | "Bottom" | First padding direction. Options: Top, Bottom, Left, Right, Left-Right, Top-Bottom. | | direction_2 | enum | "Right" | Second padding direction. |

    Outputs

    | Output | Type | What it is | |---|---|---| | image | IMAGE | Padded image tensor | | width | INT | Width after padding | | height | INT | Height after padding | | mask | MASK | Mask of the padded area (1.0 = padding, 0.0 = original image) |

    How to use it

    1. Plug your image into the image input.
    2. Set multiple_of to whatever your model needs. 8, 16, 32, 64 — depends on the model.
    3. Pick a pad_color that you can see against your image. Gray works for most things.
    4. Run your normal pipeline.
    5. After generation, crop the output back to your original size. The width and height outputs tell you the new padded dimensions so you know exactly how much to trim.

    Example

    ImagePadToMultiple example

    A 257×97 image padded to 272×112 with multiple_of=16. 15px of gray padding added on the bottom and right.

    Install

    Copy into your ComfyUI custom nodes directory and restart:

    cd ComfyUI/custom_nodes
git clone https://github.com/mr-september/comfyui-pad-to-multiple.git

    Also available through the ComfyUI Manager.

    Compatibility

    • ComfyUI: current stable
    • Python: 3.10+
    • Extra dependencies: none (just PyTorch, which you already have)

    License

    MIT

    Run ComfyUI workflows without the setup

    No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

    Learn more