ZiT t2i resolutions

ZiT - the community shorthand for Tongyi-MAI's Z-Image family - became the default local photorealism model partly because it runs on a 12GB card, and the models it runs on want specific resolutions. SBL_ZiTResT2I is a dropdown of exactly those: 22 preset resolutions organized into two tiers, 1024 and 1280, matching the model's native training buckets. Pick one, get Width and Height INTs, wire them into your resolution input, done. No hunting through a model card, no 1024x1024-everything guesswork.

Why two tiers? Z-Image's Turbo is the fast daily driver with a modest native ceiling (roughly 2MP), and the 1024 tier is its comfortable range - square at 1024×1024, then portrait and landscape variants up to 21:9 at 1344×576. The 1280 tier is the step up: same aspect ratios, bigger canvas, like 1280×1280 (1:1), 1600×896 (16:9), 1680×720 (21:9). If you're on the smaller card, 1024 tier; if you have headroom and want finer detail, 1280 tier. The aspect list is thorough - 1:1, 9:7, 4:3, 3:2, 16:9, 21:9 - so whatever shape you're composing for, there's a bucket that fits without arbitrary cropping.

How it works

Same boring-reliable mechanism as its Qwen sibling: a hardcoded map of preset label → (width, height), pure lookup, no math, no snapping. Every one of these numbers is already 8-aligned (1024, 1152, 1248, 832… all clean), so the node returns exact dimensions and you never have to remember whether ZiT wants multiples of 8 or 16. The output sockets are Width and Height (both INT), which feed Empty Latent and every modern resolution input directly.

Where it fits

If you're building a ZiT Turbo text-to-image workflow, this node is the "right resolution" answer in one click, and it pairs naturally with the SBL pack's bigger Resolution node if you want the same presets grouped inside a general-purpose 115-preset dropdown - the ZiT group there returns these identical numbers. This one just keeps the model-specific list front and center so a ZiT workflow stays tidy.

Install

As with every SBL node:

cd ComfyUI/custom_nodes
git clone https://github.com/SoftByteLabs/SBL

restart ComfyUI (or use Manager and search SBL). Zero pip dependencies, zero model downloads - it's a lookup table, so there's nothing to configure and nothing to break. One honest note: a preset is a starting canvas, not a law. When you go past first-pass generation - upscaling, outpainting, a hires pass - you'll reach for dedicated upscale nodes. For the base T2I generation, this dropdown is the whole answer.