Image to Multiband

Image to Multiband is the on-ramp. It takes a perfectly ordinary ComfyUI IMAGE - the kind every sampler and VAE spits out, shape (B, H, W, 3) - and wraps it in the pack's MULTIBAND_IMAGE container so everything downstream that speaks multiband can accept it. One input, one output, and suddenly your normal RGB image can be stacked alongside masks, feature maps, or spectral bands in a single multi-channel object.

You'll reach for this whenever the rest of your graph is built with this pack's nodes but your source is a plain image: load a checkpoint, generate, Image to Multiband, then Compose Multiband it with a mask before saving as a multi-channel file. It's the bridge between the ComfyUI you already know and the multiband world.

How it works

The IMAGE arrives as (B, H, W, 3) - height, width, then channels last, which is ComfyUI's convention. The node permutes it to (B, 3, H, W), the channel-first layout the multiband type uses internally, and labels the channels with the channel_names widget. Default is R,G,B, and that's a sensible default: keep it unless the channels mean something else in your stack.

If you supply a different comma-separated list, the node pads short lists with channel_N and truncates long ones to fit the actual channel count. So channel_names: "R,G,B" on a 4-channel image (RGBA) would give you R,G,B with the alpha channel auto-named channel_3. If you care about the alpha, name it explicitly.

The inputs

• image - required, any IMAGE tensor. This is the whole point; nothing else is needed.
• channel_names - optional, default R,G,B. Comma-separated names for your channels.

Output is a single multiband, ready for Compose Multiband, Save Multiband Image, or any other node in the pack. You can also pipe it straight back out through Multiband to Image and get your original image back - a harmless round-trip if you ever want to verify the conversion.

Installing the pack

Part of ComfyUI-Multiband. ComfyUI Manager → search "ComfyUI-Multiband" → Install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/PozzettiAndrea/ComfyUI-Multiband.git
pip install -r ComfyUI-Multiband/requirements.txt

Dependencies are numpy, torch, and tifffile. No model files - this is a pure layout conversion, nothing learned or downloaded.

One honest caveat

Don't use this node when you don't need it. If you're only ever working with normal RGB images and never stack them with anything, you've added a hop for no benefit. Image to Multiband earns its place when the image is one member of a multi-channel stack - the moment you want to save a color image alongside its segmentation masks in one file, this is the node that makes the join possible.