Quick Size (Flux)

The node that remembers Flux resolutions so you don't have to

Every Flux workflow starts the same way: you drag in an Empty Latent Image and stare at it, trying to recall whether Flux wants 1216×832 or 832×1216 for a 2:3. Quick Size (Flux) is one of five dropdown pickers in the ComfyUI-QuickSize pack, and its whole job is to turn that guessing into three dropdowns that output clean width and height integers.

Flux is a 12B flow-matching transformer that's native at 1024×1024 but genuinely happy up to about 2 megapixels - it's one of the few models where "go bigger" doesn't immediately punish you with double heads. So the tool you want isn't a fixed-size preset, it's a tier plus an aspect ratio, which is exactly how this node is shaped.

How it works

There's no API call, no model file, no math on your part - the megapixels dropdown picks one of three hardcoded lookup tables in the pack's source, preset picks an aspect key, and orientation decides which dimension comes out on top (horizontal = width is the big one, vertical = height is). Then you get two INTs to wire anywhere that eats dimensions.

The inputs that matter:

• megapixels - 1.0, 1.5, or 2.0. The author calls 1.0 the recommended tier, and it maps to Flux's classic native set: 1024×1024, 832×1216, 1152×896, 1344×768, 1536×640. 2.0 stays around 2MP (1408×1408 at 1:1, 1920×1088 at 16:9), which is squarely in Flux's comfortable range.
• preset - aspect ratio only: 1:1, 2:3, 4:3, 16:9, 21:9.
• orientation - horizontal or vertical.

Outputs are width and height as INTs. The boring but correct wiring is straight into an Empty Latent Image; anything that accepts dimensions will take them.

One honest quirk before you treat the labels as gospel: at 1:1 the 1.5 tier (1536×1536, ~2.4MP) is actually bigger than the 2.0 tier (1408×1408, ~2MP). They're tier names, not promises - Flux renders both happily, so don't lose sleep over it. And if you're the sort who insists Flux dims be multiples of 64, note that a couple of 1.5-tier values (1248×1824) miss that rule of thumb; in practice they run fine, but it's a thing to know before you chase a phantom artifact.

Installing it

Same story for all five nodes in this pack, and it's about as painless as custom nodes get - no dependencies beyond the frontend package ComfyUI already ships, no model downloads, pure Python:

cd ComfyUI/custom_nodes
git clone https://github.com/rh6423/ComfyUI-QuickSize.git

Then restart ComfyUI. Or open ComfyUI Manager, search "QuickSize", and hit install.

When it bites

The classic beginner trap here is forgetting that ComfyUI caches aggressively: change the aspect dropdown and the graph only re-runs the nodes whose inputs actually changed, so a stale preview can look like nothing happened until the sampler chain re-fires. And if you're on a 6GB card, the 2.0 tier is where VRAM goes to die - Flux at ~2MP through a GGUF or fp8 checkpoint is fine, but don't combine it with a long list of LoRAs and blame the node when the OOM hits.

It's a tiny tool from a tiny pack (rh6423, aka TaterCreekStudios on Reddit, who built it after getting "tired of typing in pixel dimensions"). The community already had Flux Resolution Calculator and Resolution Master when this shipped - this one's bet is that a per-model node with three dropdowns beats a mega-node with a model selector. For a hobbyist who swaps aspects constantly, it's a fair bet.