👁️ Preview Transparent Image

A transparent PNG is only useful if you can actually see what it looks like - and on a white canvas, transparent glass looks like a white rectangle, and on black it looks like a hole. PreviewTransparentImage is the answer: it takes your RGBA images and shows them over the background that matters, with a background_color dropdown - transparent, white, black, or checkerboard (the default). Checkerboard is the compositor's classic because it makes alpha visible in both directions: where there's transparency you see the gray squares, where there's a real edge you see it crisp against the grid.

How it works

It's an output node - images in, preview on screen, no outputs. For each input image it checks whether it has an alpha channel; if it does, it composites the RGBA over the chosen background (white or black are flat fills, checkerboard is a generated 16px grid pattern, transparent just shows the RGBA as-is on whatever the canvas gives it). If an input comes in as RGB with no alpha, it gets a full-opacity alpha slapped on so the preview doesn't crash - the node tolerates mixed input rather than dying on it. Output-wise it behaves like ComfyUI's PreviewImage: it writes to the temp/preview pipeline, so it's for eyeballing, not archiving.

Where it fits

This is the visual checkout step of the pack's transparent-Flux chain. The pipeline is FluxLayerDiffuseDecoderSimple → PreviewTransparentImage, and the whole point is to catch the classic LayerDiffuse failure modes before you commit a file: the "everything is glass" over-transparency problem (too much alpha, subject half-invisible - fix by dropping the layer LoRA strength to 0.5–0.8 per the README) and the "no alpha at all" problem (the image decoded opaque - usually the transparent VAE didn't actually load). Checking against checkerboard makes both obvious in one glance.

Inputs

• images (IMAGE, required) - the RGBA output from the decoder (or anything with alpha).
• background_color - transparent / white / black / checkerboard. Start with checkerboard; flip to white or black when you're checking how the subject will sit on the real backdrop you plan to composite onto.

That's the whole input surface. It's deliberately tiny.

The honest take

This node is the difference between "I generated a transparent image" and "I know the transparent image is actually transparent." Without a proper alpha-aware preview, ComfyUI's standard preview flattens your RGBA to RGB and you can't tell if the technique worked - which is exactly why the README's troubleshooting list includes "no transparency" as a failure mode people hit. Use it, don't skip it.

Install

Part of DenRakEiw_Nodes - ComfyUI Manager → search "DenRakEiw Nodes", or git clone https://github.com/DenRakEiw/DenRakEiw_Nodes + pip install -r requirements.txt + restart. It lives in the Flux LayerDiffuse section of the pack, so it also pulls the heavy diffusers/transformers deps whether or not you use them. And remember the surrounding truth: this whole transparency subsystem is a small author's port of a technique whose upstream tooling froze in 2024–25 - when the preview shows something wrong, the problem is usually upstream in the decode, not in this node.