Int Input

If you're serving a ComfyUI workflow as an API, this is the node you'll use for every whole number that should be callable from outside: seed, width, height, steps, batch size. Int Input is the typed member of Comfy-Pack's input family, and unlike the wildcard Any Input, it gives you real validation on the way in.

Comfy-Pack, in one line: it's BentoML's toolkit for packaging ComfyUI workflows into reproducible .cpack.zip artifacts and turning them into REST APIs. The ComfyPack input nodes define the request body of the served POST /generate endpoint, and the node's name is the API parameter name. So an Int Input node named seed means your JSON payload carries "seed": 42, and that value flows straight into whatever the node is wired to.

How it works

Mechanically it's an identity pass-through: it takes value, returns it unchanged as an INT output you can feed to a KSampler's seed slot, an Empty Latent Image's width, or any other int input. The useful part is the schema it generates when served. Comfy-Pack turns this node into an integer field with ge (greater-or-equal) and le bounds drawn from the min and max inputs. Send a seed of -5 when min is 0 and the API rejects the call with a clean validation error before the workflow burns GPU time - that's the whole point of a typed input node over an untyped one.

The inputs that matter

• value - the integer itself, default 0. This is what the API caller sets.
• min - optional lower bound, default roughly -9.2e18 (effectively unlimited). Set it to 0 for things like width/height/seed.
• max - optional upper bound, default roughly 9.2e18. Set it to something sane for steps, or leave it open.

One output: value, typed INT.

A small quirk: if min is set higher than max, the node's own validation complains - which is just the pack catching a mistake you'd otherwise make at the API boundary.

Install

Standard pack install:

• ComfyUI Manager: search comfy-pack, Install, restart, refresh.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/bentoml/comfy-pack, restart.

The pack pulls in BentoML and comfy-cli, so it's heavier than the average custom node - but it's a one-time install, and you're getting a whole workflow deployment toolkit with it, not just this one node.

The honest take

Use this for the integers you actually want to expose, and set real bounds. A served endpoint where seed, width and height are all open-range ints is a footgun waiting for a caller to send width: 9999999. Ten seconds setting min/max now saves you from explaining a 422 to someone later. And for floats - CFG scale, denoise - this node won't help; that's a job for Any Input with numeric options.