Flux Resolution Selector MXD

Flux is picky about resolution. Not in a "use exactly this or it explodes" way, but in a "models are trained on buckets, and you'll get better, more predictable output if you stick to sizes that look like the training data" way. The Flux Dev and Flux.1 Schnell training buckets are well known to the community, and Flux Resolution Selector MXD packages them into a single dropdown so you never retype 1216x832 again.

What it actually does

This is a resolution string selector, not a latent generator. Its one input is a dropdown with 18 entries - headers plus real presets - and its one output is the exact string you picked, like "Widescreen (16:9) 1344x768". The tooltip is blunt about the contract: "The selected resolution string for FluxEmptyLatentImage." It's designed to plug into the pack's own Flux Empty Latent Image MXD node, whose resolution input accepts the same strings.

Why bother with an extra hop?

Because the selector and the latent generator don't have to be in the same place. A couple of scenarios where this earns its keep:

• You want one resolution choice to drive several empty-latent nodes - a main latent, a hires pass, an inpaint - without editing three dropdowns.
• You're building a workflow template where resolution is meant to be moved around by the user without touching the hidden plumbing.
• You're wiring a "resolution" input to a front-end or a preset system and want a stable string interface.

The dropdown groups are sensible: High (1408² up to 2176x960), Standard (the 1024² class), and Low (320² down to 576x256) for quick previews. All of them are multiples of 64/32-friendly sizes that land cleanly in Flux's latent space.

The honest caveat

If you're hand-building a one-off Flux workflow, you don't need this node. Just set the resolution directly on Flux Empty Latent Image MXD and move on - the pack already bundles the presets there. The selector earns its place when you're standardizing, sharing, or building tooling around a workflow. It's a thin node, and it knows it.

Install and requirements

Same story as every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

or ComfyUI Manager → search "Maxed Out" → install → restart. No model downloads, no extra pip packages, and it doesn't touch the model at all - it just passes a string through.

Troubleshooting

• "… is a header or invalid option" - if you feed the selector's output somewhere that expects an actual resolution and you picked a header row (the - High Resolutions - entries), downstream will choke. Headers are selectable in the dropdown but aren't real resolutions; keep your pick on an actual preset.
• Feeding it into the wrong node - the output is a string. If you plug it into a node that wants an INT width/height (like the core Empty Latent Image), it'll fail validation. It's built for the MaxedOut empty-latent nodes that speak the same vocabulary.

For a piece of workflow plumbing it's genuinely handy - small, predictable, and it removes one more place where typos creep in.