ITL Resolution Selector

Every diffusion model has resolution rules, and almost none of them are the same. SD 1.5-era models want multiples of 64, most samplers tolerate multiples of 8, and Ideogram 4 insists on multiples of 16 within a 256–2048 pixel range per side. Type 1025×1024 into an Empty Latent and you've already handed the model a subtly broken input. The ITL Resolution Selector exists so you pick an aspect ratio and a target size, and get back width/height that actually satisfy whatever model profile you're running.

It's a plumbing node - two INT outputs, zero image processing - but it's the kind of plumbing that quietly prevents a whole class of "why does my image look off" problems.

How it works

You pick a profile, a resolution mode, and an aspect ratio, and the node computes width/height snapped to the profile's multiple. Three things drive it:

• profile - default does no clamping, it just snaps each side to snap_multiple (default 8, which is what most diffusion samplers need). The Ideogram 4 profile is stricter: multiples of 16, clamped to 256–2048 px per side, and it shows a "clamped to keep aspect" warning when a request exceeds the cap. default is the default so the node is safe to drop into any workflow.
• resolution_mode - raw (type width/height directly), auto (pick a ratio, edit one side, the other follows), or megapixel (target megapixels + ratio, both sides computed). Megapixel is the default and the one most people end up using.
• aspect_ratio + orientation - named presets from 1:1 Square out to 3:1 Wide Panorama, including 21:9 Cinemascope. The list is square + landscape only; the ⟷ flip button transposes to portrait (16:9 → 9:16), so there's no duplicated 1:1.

The readout shows the resulting W × H, the megapixel count, and the effective ratio + orientation, and the math is mirrored between the Python node and the front-end JS so the number you see is the number that runs.

What you actually touch

• snap_multiple (default 8) - leave it alone unless your model needs a different step. Model profiles like Ideogram 4 override it with their own multiple anyway.
• megapixels - the target size; note Ideogram 4 still clamps at ~4.19 MP (2048×2048), so cranking this past 4 won't help with that profile.
• width / height - editable in raw and auto mode, computed in megapixel mode.

Outputs are width and height (both INT). The obvious wiring: into an Empty Latent (Image) for most models, or into the ITL Ideogram 4 Prompt Builder's width/height inputs, where edits push into the builder's canvas live and also apply on execution.

Installing it

ComfyUI Manager (search "ComfyUI-IntoTheLatent-Utils") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Into-The-Latent/ComfyUI-IntoTheLatent-Utils

then restart. No extra dependencies beyond what ComfyUI already ships.

Is this node essential? No - if you're only ever generating 1024×1024 you don't need it. But the moment you're doing banners, posters, or phone-wallpaper crops, a ratio-driven picker that knows Ideogram 4's rules beats doing the math in your head. Just remember the pack ships no support (issues are closed on the repo), so if the clamp behavior surprises you, the answer is in the tooltip.