WAN2.2 Empty Latent Video MXD

Core ComfyUI ships Empty Hunyuan Latent Video - the standard way to get an empty video latent so you can noise it up and sample. It's useful, but its width/height boxes are a blank slate, and if you've ever sat there doing the WAN 2.2 bucket math by hand ("832x480, that's 104x60 in latent space, t is (81-1)/4+1 = 21..."), you know why that's a problem. WAN2.2 Empty Latent Video MXD is the same node, rebuilt around WAN 2.2's actual resolution buckets, with a vertical toggle so you never do division in your head again.

What it is

A preset-driven empty video latent generator. Pick a resolution from the dropdown, set length and batch_size, and you get a LATENT wired straight into your WAN 2.2 sampler. It's part of the MaxedOut pack's WAN 2.2 toolkit, aimed at the same people running the author's Patreon video workflows.

The presets cover the two tiers WAN 2.2 was trained around:

• 720p: Widescreen 1280x720, Square 1024x1024
• 480p: Widescreen 832x480, Square 624x624

That's it, and honestly that's enough - these are the buckets that make WAN 2.2 sing, and anything outside them is where the quality falls off.

How it works

The math is identical to core EmptyHunyuanLatentVideo, just with presets:

• Spatial: [h//8, w//8] - WAN's VAE compresses 8x.
• Temporal: t = (length-1)//4 + 1 - 4x temporal compression, so 81 frames becomes a 21-frame latent.
• Channels: 16, WAN's latent channel count.
• vertical swaps width and height first.

You end up with a [batch, 16, t, h//8, w//8] zeros tensor wrapped as a latent, ready for noise and sampling. The default length of 81 = 5 seconds at 16fps, which is the classic WAN clip length.

Inputs and outputs

• resolution - the dropdown, default Widescreen (16:9) 832×480 (the 480p default - the sensible starting point on most consumer GPUs).
• vertical - swap to portrait.
• length - frames, default 81, steps of 4 to stay aligned with the 4x temporal compression.
• batch_size - 1 by default.
• Output: LATENT - into your WAN 2.2 sampler's latent input.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Restart, or ComfyUI Manager → search "Maxed Out" → install. No extra pip deps; you'll still need your WAN 2.2 checkpoint and VAE.

Troubleshooting

• "… is not a selectable resolution" - you picked a section header (- 720p -). Choose an actual preset.
• Sampler errors about latent shape - you're mixing latent families. This node makes WAN-shaped latents; if you're sampling a different model it'll reject them. Match the node to the model.
• OOM at 720p square - 1024x1024 video is heavy on VRAM. Drop to the 480p buckets; WAN still looks good there and you'll actually finish a clip.

For "I just want a correct empty WAN video latent", this is the fastest path in the pack. Pick, run, done.