S1Z35 - Flux 1

Flux doesn't need the same resolution babysitting SDXL did - it's happy up to around two megapixels, and its T5 encoder makes prompting a lot less fragile. But it still rewards starting from sizes it actually trained on, and if you build Flux workflows more than once, S1Z35 - Flux 1 is the little node that keeps you from typing 1344×768 for the hundredth time.

What it is

A size-preset node from the ComfyUI-S1Z35 pack, for FLUX.1. Pick a preset, get width and height as integers, wire them into your latent. The flux1.csv ships with the standard eight buckets - square 1024, 4:3, 3:2, 16:9, 9:16 and friends, both landscape and portrait. That's the same classic set as the pack's SDXL node, which makes sense: these are bucket sizes the model saw during multi-aspect training, not numbers someone made up.

FLUX.1 is worth the presets for one specific reason: it's a 12B rectified flow transformer, the model that replaced SDXL as the open-weights quality bar back in 2024, and it's still the baseline newer models get measured against. It's also a VRAM hog in its full fp16 stack, which is why most people run it quantized - fp8, GGUF, NF4. A resolution dropdown doesn't fix any of that, but it does remove one more fiddly thing from a workflow that's already carrying quantization nodes.

How it works

Same shared engine as the rest of the pack. At node creation it reads presets/flux1.csv, turns each width, height, label line into a dropdown entry like Cinematic (16:9) (1344x768), and on selection parses the numbers back out. No model loading, no dependencies, no GPU work - pure stdlib CSV parsing.

Inputs and outputs that matter

Two inputs, both simple:

• preset - the dropdown. The only thing you'll normally touch.
• swap_dimensions - boolean, off by default. Flips width and height, so Landscape (3:2) becomes portrait without any file editing.

Outputs are width (INT) and height (INT). The classic wiring is into EmptyLatentImage's size inputs before your KSampler. If you're running a hires-fix or a two-pass setup, you can also grab these to drive your latent resolution and let the upscale stage handle the rest.

Installing it

The pack is one of the lighter installs you'll do - no requirements.txt, no model downloads, no config:

cd ComfyUI/custom_nodes
git clone https://github.com/K3nt3L/ComfyUI-S1Z35

Restart ComfyUI and it appears under S1Z35/Resolution. Or use ComfyUI Manager → Install Custom Nodes and search for "S1Z35".

Common issues

• Your edited presets don't appear. The dropdown is built when the node is created, so after editing presets/flux1.csv you need to delete the node and add it fresh (or restart) to see the changes.
• "Missing flux1.csv (512x512)" is the shipped fallback when the file is absent or empty - and it will happily give you a 512×512 latent, which Flux barely tolerates. If you see it, your CSV is gone or unreadable.
• Don't push past the list just because Flux can. FLUX.1 generates comfortably in the two-megapixel range, but this node's presets stop at 1536×640, and that's a feature - going huge without upscaling is where artifacts creep in. If you want 1536×1536, add it to the CSV in width, height, label format; the parser will pick it up on the next fresh node.

Keep in mind Flux runs on guidance scale rather than CFG, and a wrong scheduler choice shows up as muddy output - but that's Flux, not this node. The node just hands you sizes, and the sizes it ships with are the ones Flux likes.