DCI Preview

You can't view a .dci binary archive in a normal image viewer - it's a container holding a dozen differently-named icon layers. So before you package and ship, you want this node: DCI Preview takes a BINARY_DATA DCI blob and renders it as a visual grid, splitting the icons into Light and Dark columns and overlaying the metadata you actually care about.

It's the "is this thing right?" checkpoint of the whole pack. Feed it the output of a DCI File node (or a loaded .dci file from the Binary File Loader), run, and you get a montage of every image inside, side by side - light tone on one side, dark tone on the other, each tile labeled with its size, state, scale, and path. If you're building a theme, this is how you catch "hover is the same as normal" before you've committed it to a package.

Inputs are minimal:

• DCI Binary Data - the required input, and it accepts either a single blob or a list of blobs. Feed it the whole batch from a Directory Loader or Deb Loader and it'll render each one.
• Light Background Color and Dark Background Color - dropdowns with 20 options (Transparent, White, Black, Checkerboard, Light Gray, Dark Gray, the named colors like Blue/Red/Cyan…). The default of Light Gray / Dark Gray is a good neutral pairing: each side gets a background that makes icons of that tone readable.
• Text Font Size - 8–50, default 18. Controls the metadata labels on the grid.

Output is a single Preview Images tensor (IMAGE), which you can save with a standard Save Image node or wire onward. The node is also an output node - it shows the grid in the UI automatically, complete with a text summary of what it found. And it's unusually good at explaining failure: instead of a dead silence, it renders a red error image with the reason written on it (missing data, unreadable format, no images found), plus a detailed message in the console.

Installing

Same pack, plain install - Pillow and numpy only, no models:

cd ComfyUI/custom_nodes
git clone https://github.com/zccrs/comfyui-dci
cd comfyui-dci && pip install -r requirements.txt

Or "DCI Image Export Extension" via ComfyUI Manager, then restart. Nodes live under DCI.

Where people get burned

• It needs the DCI container, not a single image. If you're previewing one layer, use the sibling DCI Image Preview node instead - this one expects a full archive (or a list of them) and will complain if it can't parse the input as DCI.
• Checkerboard is applied as a compositing trick, not real alpha. The preview simulates transparency over a checkerboard; the underlying icon keeps whatever alpha it had. Good enough to eyeball, not a pixel-perfect alpha audit.
• List inputs stack. Feed it a list of archives and you get a batch of grids, not one merged image. That's usually what you want - each file gets its own grid.

If the DCI file is valid, this is the fastest way to feel like you actually understand what you built. When it isn't valid, it's the fastest way to find out why - the error previews do the debugging for you.