Gemini FLUX Resolutions

First, clear up the name: this node does not call Gemini and needs no API key. It just carries the pack's "Gemini" prefix because it ships in ComfyUI-OllamaGemini. What it actually is: a resolution picker. You choose a labeled image size from a dropdown - "1024x1024 (1:1) - Square Large", "1366x768 (16:9) - Laptop Standard", "1536x768 (2:1) - Cinematic Wide" - and it hands you two integers, a width and a height, to feed your Empty Latent Image.

If that sounds identical to the pack's plain FLUX Resolutions node, it nearly is. The difference is the menu: this one carries a longer list (forty presets versus thirty), with a few extra wide and cinematic buckets like the 16:9 laptop size. Pick whichever node has the ratio you want in its dropdown; there's no deeper distinction.

Why you'd bother

Resolution is the setting beginners get wrong most often, and it punishes you loudly. Diffusion models are trained on a specific pixel budget across a set of aspect-ratio buckets - SDXL around 1024, FLUX a bit more forgiving but still happiest near what it saw. Ask for a size well outside that and the model doesn't politely rescale; it duplicates the subject, clones heads, or stretches bodies to fill the odd canvas. This dropdown is just forty pairs of numbers that all sit in the safe zone, spelled out so you don't have to remember which are blessed.

How it works

Every menu entry is a hard-coded width/height pair. Select one, the node emits those two numbers. That's the whole mechanism - a lookup table with friendly labels.

Two extra knobs:

• multiply_factor (default 1) - multiplies the picked dimensions by a whole number. A 2 turns 1024x1024 into 2048x2048. Useful for planning, but it is not an upscaler; generating natively at 2x reintroduces the very artifacts the presets avoid.
• manual_width / manual_height (default 0) - override. At 0 the dropdown wins; enter real numbers and they replace the preset, for when you need an exact custom size.

Inputs and outputs that matter

You really only set size_selected. The outputs are width and height, both integers, and they wire straight into the width/height inputs of an Empty Latent Image node (or anything that takes dimensions). The node sizes the canvas; your sampler paints on it.

Installing it

ComfyUI Manager is the easy path: search ComfyUI-OllamaGemini, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/al-swaiti/ComfyUI-OllamaGemini
pip install -r requirements.txt

and restart. The node itself pulls no models and needs no keys, but you install the full pack to get it. If the pack's dependencies fail to install, the whole thing won't import and the node simply won't appear - check the startup console for import errors before assuming it's gone.

Where people get burned

Same two traps as its twin. multiply_factor looks like free resolution and isn't - crank it and you're generating at a size the model dislikes, so duplicate-subject and stretch artifacts come back. Generate sane, upscale separately. And the width/height are dumb integers with no knowledge of your checkpoint, so if the aspect ratio fights the model's preference, that's on your wiring, not the node. If you're deciding between this and the plain FLUX Resolutions node, just open both dropdowns and use whichever lists the ratio you actually want.