WanResolutions

Every Wan workflow ends at the same two integers - width and height - and most people get there by typing 1280×720 into a box and hoping. WanResolutions replaces the hoping with a dropdown that understands Wan 2.2's actual alignment requirements. No model downloads, no API key, nothing to fetch: it's a picker that outputs the two INT values your Wan 2.2 A14B workflow actually wants.

Why this matters more than it sounds: Wan 2.2 has a hard resolution grid. Its VAE downsamples 8× per side, and the DiT patches 2× on top, so every dimension has to be divisible by 16 to be legal. Generate off-grid and you get smeared edges, glitches, or a plain refusal to run. That's the entire problem this pack exists to delete - every size it emits is divisible by 16, so there's no rounding flag to remember and no "did I do the math" moment.

How it works

Pick an aspect_ratio (1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9) and a resolution tier. There are four tiers per aspect: Fast Draft, Preview, High Detail, and Wan 2.2 Native. The first three are lower-area conveniences for iterating fast - draft a clip at 480×480, check motion and composition, then jump to the real thing. "Wan 2.2 Native" is the one that matches what Alibaba's own pipeline targets, so for 16:9 that's the familiar 1280×720. Switch aspects and the node keeps you on the same quality tier rather than resetting you to the bottom, which is the small quality-of-life touch that makes it feel designed rather than scripted.

The two inputs you'll actually touch:

• aspect_ratio - landscape/portrait/square. Default 1:1.
• resolution - the tier dropdown. It's seeded with the current aspect's tiers and swaps as you change aspect, so you're never staring at a 60-item list.
• official_only (default off) - Wan's own CLI only ships four buckets: 1280×720, 720×1280, 832×480, 480×832. Flip this on and the dropdown swaps to just the official sizes for your current orientation (480P / 720P), with the tier snapped to the nearest bucket by area. Useful when you want outputs that match Wan's reference implementation exactly.

There's also an optional IMAGE input: connect a frame and the node picks the closest supported aspect ratio for you. Nice for I2V, where the source clip already tells you the shape.

Outputs are exactly two: width and height as INTs. Wire them into whatever your Wan setup takes resolution from - the Wan loader/empty-latent stage in your workflow, or Kijai's WanVideoWrapper if that's the path you're on. That's the whole node.

Installing

ComfyUI Manager is the easy route - search "WanResolutions" (or "Wan/MiniMax/LTX Resolutions", the pack's display name) and install. Or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/boobkake22/ComfyUI-WanResolutions.git

Restart ComfyUI after either one. There are no Python dependencies (the source is pure stdlib) and no model files to download - the pack is Apache-2.0 and about as harmless as a custom node gets.

Where people get burned

• Connected an image and the aspect seems stuck? The IMAGE input overrides your manual selection by design. If you want manual control, leave the image unconnected.
• official_only made your dropdown tiny? That's the feature. It intentionally hides the non-official tiers so you can see what's being filtered out.
• Old workflows still load? Also by design - the node parses legacy labels from before its preset renames, so saved workflows don't break when the pack updates.

Is it glamorous? No. But for Wan 2.2 it's the difference between "did I pick a legal resolution" and actually making video. For a workflow that already costs minutes per clip, that's a real trade.