Quick Size (Wan)

The node that keeps a wrong Wan aspect from costing you a long render

Wan is Alibaba's open video family - the 14B and 5B diffusion transformers that became the base layer for most local video work, frozen at 2.2 but still everywhere. Video is the worst place to guess resolutions: a wrong aspect means wasting a multi-minute, VRAM-hungry render on a stretched mess, and the author of ComfyUI-QuickSize put it bluntly in the pack's announcement - every extra click is "a typo waiting to wreck a 10 minute video gen." Quick Size (Wan) is the pack's answer: dropdowns that output width and height integers for Wan's actual 480p and 720p grids.

This is the pack's most configurable node, and it needs to be, because video has two dimensions of choice image nodes don't. The inputs that matter:

• model_size - Wan 14B or Wan 5B. This matters, not just as a label: the 5B's 720p grid uses a 708px short side to match its native downsampling, while the 14B uses the standard 1280×720. Feed a 5B workflow the 14B's numbers and you can get artifacts.
• video_size - 720p or 480p. The 480p tables are identical for both models (832×480 at 16:9, 640×480 at 4:3, 576×576 at 1:1); the 720p tables are where they diverge.
• preset - nine aspects: 1:1, 3:2/2:3, 4:3/3:4, 16:9/9:16, and 21:9/9:21. Most video lives at 16:9, and it's the default.
• orientation - horizontal or vertical.

Outputs are width and height INTs, wired into Wan's latent/empty-latent setup or any node that wants video dimensions.

Two gotchas worth knowing, and both are grounded in how people actually run Wan. First, the 5B's 708px short side is a deliberate-but-unusual choice - it isn't a multiple of 8, and Wan work generally wants dimensions on a clean grid. A commenter in the pack's announcement thread recommended routing Wan sizes through a resize node set to multiples of 8; if a downstream node complains or you see seam artifacts at the 5B/720p presets, snap 708 up or down to 704 or 720. Second, the model_size dropdown and your actual checkpoint must agree - it's a table selector, not a detector, and it will happily hand you 1280×720 for a 5B workflow if that's what you tell it.

Installing it

Same as the rest of the pack - featherweight, no model files, no extra Python dependencies:

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

Restart ComfyUI. Or ComfyUI Manager → search "QuickSize" → install.

Verdict

For image work this pack competes with heavier tools like Resolution Master, but for video it's the clear lightweight option - Wan's grids are fiddly enough that even people who roll their own size nodes end up reimplementing exactly this. It won't save you from a bad prompt or a 14B OOM, but it will save you from the special kind of heartbreak where you realize ten minutes in that the video is 4:3. Small pack, small author, genuinely useful niche.