ComfyUI-ResizeDimensions

ComfyUI-ResizeDimensions

A single ComfyUI custom node — Resize Dimensions (W/H) — that computes latent-safe width/height from an image's aspect ratio (or fixed sizes) and can also output a matching empty LATENT batch, so you can skip a separate Empty Latent Image node. It does not resample the image; the image is read only for its aspect ratio.

Install

Via ComfyUI-Manager: Install via Git URL → paste this repo's URL, then restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/normbarrette-arch/ComfyUI-ResizeDimensions.git

Restart ComfyUI. The node appears under image/size as Resize Dimensions (W/H).

Node: Resize Dimensions (W/H)

A mode selector chooses one of three sizing functions; only the active one is output. Both outputs are snapped to round_to.

Inputs

• image — source image (only its aspect ratio is read, for the edge modes).
• mode:
  • long edge — pin the longer source side to edge_length; the other side scales (aspect preserved).
  • short edge — pin the shorter source side to edge_length; the other side scales (aspect preserved).
  • set size A — output exactly width_a × height_a.
  • set size B — output exactly width_b × height_b.
• edge_length — target for the long/short edge modes.
• width_a / height_a, width_b / height_b — exact sizes for the set-size modes.
• round_to — snap both outputs to this multiple (8/16/32/64/1).
• latent_type — latent channel count for the LATENT output: SD3 / Flux / Qwen (16ch) or SDXL / SD1.5 (4ch). Must match your model.
• batch_size — number of empty latents in the output batch.

Outputs

• width (INT), height (INT) — the single selected pair.
• latent (LATENT) — an empty (zeros) batch at the computed size. Leave it unconnected if you only need width/height.

Every output is snapped to round_to, so even a typed set size value is nudged to the nearest multiple — set round_to = 1 for an exact odd value (note the latent floor-divides by 8, so keep round_to at 8+ for an exact size match).

License

MIT — see LICENSE.