Video Resolution

Video models are picky about resolution in a way image models aren't. A video pipeline wants a consistent, model-appropriate resolution threaded through the latent creation, the sampler, and the output settings - and it wants it as plain numbers. Video Resolution is the settings node that turns "I want 9:16 phone-style footage" into a clean width and height pair you can wire into the rest of the graph without doing arithmetic in your head.

What it does

It's a resolution selector with a dropdown of presets and a Custom escape hatch. Pick a preset and it emits the matching width and height as integer outputs; pick "Custom" and it uses the width and height widgets instead. That's the whole node, and it's deliberately dumb - the value is in the preset list being correct and consistent.

The presets cover the two orientations you actually see in video work:

• 9:16 (portrait) - the phone/reels format: 240p through 8K, with 576x1024 as the default and 720x1280 (HD) the one most people actually land on.
• 16:9 (landscape) - 1024x576 default, up to 7680x4320.

Plus a couple of bare ratio entries like 480x832 if you want to eyeball raw pixel counts.

Where it wires in

The two integer outputs, width and height, feed anything that needs dimensions: an Empty Latent node for a video-size latent, the Eclipse Context (Video) or WanVideo Setup pipes, or a resolution widget elsewhere in the graph. The point of a node like this over just typing numbers into the latent node is that the resolution is defined once, in one visible place, and shared - change it in one spot and every consumer updates. That's the same one-source-of-truth instinct behind the whole Eclipse settings family.

For WAN and most modern video models the practical pattern is: pick the preset that matches your model's training sweet spot (WAN 2.x is happy around 576x1024 for fast runs, 720x1280 when you have VRAM), wire the outputs into your context pipe, and let the pipeline stay internally consistent. An 832x480 latent with a 720x1280 video context is how you get weird letterboxing and shape mismatches that are a pain to chase down.

Install it

Part of ComfyUI_Eclipse. ComfyUI Manager → search "Eclipse" → Install, restart, and it's under Eclipse → Settings:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI_Eclipse

Standard pack deps (opencv-python, pilgram, PyYAML, aiohttp; portable installs may need pip install -r custom_nodes/ComfyUI_Eclipse/requirements.txt).

Gotchas

Two things. First, "Custom" is sticky - if you set custom dimensions once, the node keeps using them until you switch the dropdown back to a preset, so a workflow you "fixed" by choosing a preset can silently revert to your old custom numbers. Second, remember that most video VAEs need dimensions divisible by a power of two (16 or 32 depending on the model); the presets are already aligned, but if you go Custom, don't type 777x1024 and expect the sampler to shrug it off. Check your model's preferred multiple, then round to it.