BitDance Resolution

This is the smallest node in the pack and also the easiest to skip - right up until your sampler throws "Unsupported image size" at you. BitDance doesn't generate at whatever dimensions you feel like. The model was trained on a fixed list of 18 resolutions, and BitDance Sampler validates your input against that exact list. BitDance Resolution exists to hand the sampler one of those valid sizes in the format it expects, instead of you guessing and praying.

The node is a single dropdown - resolution, defaulting to 1024x1024 - and it gives you four outputs for the price of one:

• resolution (BITDANCE_RESOLUTION) - the object you wire into the sampler's resolution input. This is the one that matters.
• width / height (INT) - the pixel dimensions, if you want to feed them to a size-dependent node elsewhere.
• label (STRING) - the WxH text, handy for filenames or logging.

The preset list

The 18 choices run from 512x2048 (portrait) through 2048x512 (ultrawide landscape), with 1024x1024 at the center and a handful of small sizes like 256x1024 and 512x512 down at the bottom. Everything in the dropdown is a real, model-validated size - that's the point. If you've been burned by other models where the "good" resolution is a folk rumor, here it's enforced in code: the sampler checks your selection against the same internal list this node draws from, and a mismatch is a hard error, not a soft quality hit.

How to actually use it

Drop it into the graph, pick an aspect ratio, and wire the resolution output into the sampler. That's it. The width/height and label outputs are there for flexibility, but the minimal workflow doesn't touch them.

Two practical notes. First, bigger is not automatically better: BitDance's sampling cost scales with total token count, so a 2048x512 image costs roughly the same as a 1024x1024 (same total pixels), and the 16–24 GB VRAM crowd will have an easier time with the 1024-class presets. Second, the README's own fix guide for the "generic face" issue leans on resolution choices too - faces at small per-face pixel counts look generic in any model, and picking a preset that gives your subject room is part of the fix.

Install with the rest of the pack: ComfyUI Manager → search "Comfyui-bitdance", or git clone https://github.com/aistudynow/Comfyui-bitdance into custom_nodes + pip install -r requirements.txt + restart. There's nothing else to configure - the whole node is one dropdown and four outputs, which is exactly as much as it needs to be.