DCI Image Preview

If DCI_PreviewNode is the auditor that reviews the whole archive, this is the designer's mirror - the one you drop in while you're still shaping a single icon. DCI Image Preview takes one (or a list of) DCI_IMAGE_DATA packets straight from the DCI Image / DCI Sample Image nodes and shows you the actual rendered layer, over whatever background you pick.

The key difference from its sibling: it eats the intermediate DCI_IMAGE_DATA format, not a finished .dci binary. So you can chain it directly off an image node before you've even assembled the container - check that the resize looks right, that the transparency holds, that the layer padding isn't eating the artwork - then continue on to the DCI File node with confidence.

Inputs:

• DCI Image Data - required; accepts a single packet or a list of them.
• Preview Background - the one knob: Transparent, White, Black, or Checkerboard (the default). Checkerboard is the classic choice for icons since it makes alpha edges visible; White and Black let you sanity-check how the icon reads on an actual theme background.

Output is a Preview Images tensor (IMAGE), savable with a regular Save Image node. It's also an output node, so the preview shows in the UI automatically, with a small text label describing the path, dimensions, and color mode. Two nice touches in the source: if the layer has padding set, it draws a dashed border showing the content box, and if it can't render for any reason it produces a red error image instead of failing silently.

Installing

It's part of comfyui-dci - a light install, just Pillow and numpy:

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

Or search "DCI Image Export Extension" in ComfyUI Manager and restart. All DCI nodes are under the DCI menu category.

Where people get burned

• It's for layers, not archives. If you connect a full .dci BINARY_DATA here, it won't parse - that's the DCI_PreviewNode's job. Match the input format to the node.
• Padding visualization can mislead. The dashed border converts padding to pixels heuristically; treat it as a visual aid, not a measurement.
• RGBA previews over "Transparent" rely on the checkerboard trick elsewhere in the pack - if you want a rigorous alpha check, render over White and Black and compare.

Honestly, it's a small node with a small job, and that's fine. One input, one output, one knob. If you're iterating on a single icon and want to see it immediately without committing to a full archive, this is the one you reach for.