ComfyUI Extension: ComfyUI-WanResolutions
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.
ComfyUI custom node for selecting WAN resolution presets by aspect ratio.
README
ComfyUI WanResolutions
ComfyUI custom nodes that output width and height presets for Wan 2.2 and LTX video generation.
Nodes
WanResolutions
- Lets you choose an
aspect_ratioand resolution quality tier (Fast Draft,Preview,High Detail,Wan 2.2 Native) - Outputs
INTvalues:width,height - All sizes are divisible by 16 — the real alignment requirement for the Wan 2.2 A14B T2V/I2V models (8× VAE stride × 2× DiT patch), so no rounding flag is needed
Wan 2.2 Nativematches what the model actually targets (e.g. 16:9 →1280×720); the lower tiers are faster, lower-area conveniences- Optional
IMAGEinput auto-matches the closest supported aspect ratio as an override - Optional
official_onlyrestricts output to Wan's four official CLI buckets (1280×720/720×1280/832×480/480×832). Enabling it swaps the resolution dropdown to show just the two official sizes for the current orientation (480P/720P), so it's clear what is being filtered out; the current tier is mapped to the nearest bucket by area - Keeps equivalent quality tier when you switch aspect ratios in the UI
LTXResolutions
- Uses the same aspect-ratio workflow as
WanResolutions - Uses dimensions divisible by
32 - Adds
image_bypassso a connectedIMAGEinput can be ignored when needed - Adds
upscaler_poweroptions fornone,x1.5, andx2; upscale modes output the lower base resolution needed to reach the selected target tier - The
upscaler_powercombo accepts a direct connection from theLTX Upscaler Powersupport node or any compatible generic/combo switch - Preset tiers are tuned around current LTX 2.3 guidance:
- official fast-iteration aspect sizes such as
640×640,512×768,768×512,512×704, and704×512 - the documented two-stage workflow, where lower preview passes feed higher output tiers
- the current official landscape workflow pair of
960×544base resolution and1920×1088x2 output 720p/1080p-class output targets, snapped to legal multiples of32and adapted per aspect ratio
- official fast-iteration aspect sizes such as
LTX Upscaler Power
- Outputs a connectable
COMBOvalue:none,x1.5, orx2 - Connects directly to the
LTXResolutions.upscaler_powercombo, or can pass through a compatible generic/combo switch first
Supported aspect ratios for both nodes:
1:12:33:23:44:39:1616:9
Installation
Clone into your ComfyUI custom_nodes directory:
cd ComfyUI/custom_nodes
git clone https://github.com/boobkake22/ComfyUI-WanResolutions.git
Restart ComfyUI after installing.
Usage
- Add either
WanResolutionsorLTXResolutions. - Pick
aspect_ratioandresolution. - Optional: for
WanResolutions, enableofficial_onlyto force one of Wan's four official buckets. - Optional: connect an
IMAGEinput to auto-select the nearest aspect ratio. - Optional: on
LTXResolutions, enableimage_bypassto ignore the connectedIMAGE. - Optional: select an
upscaler_power, or connectLTX Upscaler Powerdirectly to theupscaler_powercombo. - Connect
widthandheightoutputs to your downstream nodes.
Notes
- The LTX presets follow current LTX 2.3 notes rather than older
0.9.xguidance. The documented anchor sizes are640×640,512×768,768×512,512×704, and704×512. Current official ComfyUI workflows use960×544before x2 spatial upscaling and1920×1088afterward. - Upscaler modes treat the selected preset as the desired final target, round it upward to a factor-compatible legal size when necessary, and output a base resolution divisible by
32. WanResolutionssizes are derived from Wan 2.2 A14B's true 16-pixel alignment grid: dense aspect ratios (1:1,2:3,3:2,3:4,4:3) land on exact ratios, and16:9/9:16sit within ~0.6% (sub-pixel) while keeping an even area ladder.Wan 2.2 Nativereproduces the sizes Wan's own pipeline emits.
License
Apache-2.0. See LICENSE.
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.