Bucket Ratio List

What it's for

This is a reference card with a node attached. One widget, one string out. It exists because the pack's real node - Bucket Ratio Crop - picks a bucket on your behalf, and "which buckets were even on the menu?" is a question you ask the moment a crop comes back at a size you didn't expect.

No image, no model, no VRAM. It runs instantly whether or not anything upstream of it exists.

How it works

Pure lookup. You pick a group, it formats the table that BucketRatioCrop chooses from, one line per bucket, with aspect ratio and pixel count. That's the entire node - there's no detection and no cleverness.

Three groups, and the numbers are the useful part. Every bucket inside a group has both sides as multiples of 64 and a fixed width + height: 1024 for the 512 group, 1536 for 768, 2048 for 1024. The group number is the square bucket, and the square has the most pixels in its group - pushing toward an extreme ratio trades area away for length.

Bucket group 512 (13 ratios):
   128 x 896   ratio=0.14  pixels=114688
   192 x 832   ratio=0.23  pixels=159744
   256 x 768   ratio=0.33  pixels=196608
   ...
   512 x 512   ratio=1.00  pixels=262144
   ...
   896 x 128   ratio=7.00  pixels=114688

| Group | Ratios | Pixels per bucket | |---|---|---| | 512 | 13 | 0.11 – 0.26 MP | | 768 | 17 | 0.33 – 0.59 MP | | 1024 | 25 | 0.46 – 1.05 MP |

Note what that last column does to the group name. 1024 does not mean "1 megapixel" - only 1024 x 1024 hits that. A tall 256 x 1792 is in the same group at less than half the pixels. The group is a budget, not a resolution.

The input and the output

One input: bucket_group, the same 512 / 768 / 1024 dropdown the crop node has.

One output: bucket_info, a STRING - a multi-line text blob. It's not an image, not a latent, nothing the sampler can eat. Wire it into any node that takes text for display: a Show Text / Preview Text node, a text concat feeding your save filename, a note you screenshot for your own workflow. If you've ever wanted to know whether 1024 can give you a 7:4 crop before building the graph around it, this answers it in one click.

Installing it

Same pack as the crop node, so if you already have one you have both. ComfyUI Manager → search BucketCrop → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/malcolmamal/ComfyUI-BucketCrop
# restart ComfyUI

No dependencies to pip install, nothing to download - the pack's pyproject.toml declares no requirements and there's no model loading anywhere in it. The README is an empty file, so don't go looking there for documentation.

Rough edges

Not many, because there's not much to break. Two real ones:

It knows nothing about your image. It's a lookup table, so it will happily list all 25 ratios of the 1024 group while the photo you're actually cropping is too small to reach most of them. BucketRatioCrop filters to buckets that fit your source; this node doesn't know that filter exists.

The 1024 group's output is 25 lines. Into a text box that's a scroll, not a view, and if you're feeding it to a string-length-limited widget you'll get a truncated table. Use a multiline display node, or just screenshot it once and move on.

That's genuinely it. Treat it as documentation that lives inside the graph - check it when you're building a training-dataset crop and forget about it after.