Hunyuan 3D 2.1 Load Image with Transparency

Stock ComfyUI's Load Image flattens your PNG to RGB and the alpha channel quietly ceases to exist. Fine for image work. Annoying for 3D, where a transparent-background render is the whole point - you want the object, not the background it was saved on. Hy3D21LoadImageWithTransparency is the version that keeps the alpha: same input-picker UI as Load Image, three outputs instead of two.

What you get out

• image - the standard RGB tensor. This is what you feed into a Hunyuan 3D 2.1 generator node, same as you'd feed any ComfyUI image.
• mask - the alpha channel, converted to a ComfyUI MASK (inverted so white = opaque object, matching ComfyUI convention). Handy if you want the silhouette for anything downstream - background-aware workflows, compositing, that sort of thing.
• image_with_alpha - the original image tensor before the RGB flatten, alpha intact. The "with transparency" output the node is named for.

It also behaves like the node it replaces: animated formats (GIF/WebP) are iterated into a batch, EXIF orientation is corrected, and the node tracks file changes by content hash so ComfyUI doesn't cache-stall on a re-saved image.

Why this node exists

Honestly? Mostly for the pack's own workflows. When you're generating a 3D mesh from a transparent product shot, losing the alpha in the first step means you're either re-cropping by hand or relying on the pipeline's background remover later. This node hands the generator a clean RGB frame and keeps the transparency around as a mask so the whole graph can make use of it. If you're not doing 3D, the standard Load Image does everything this does except the two extra outputs - so don't reach for it unless the alpha matters.

Using it

It's part of the ComfyUI-Hunyuan3d-2-1 pack, so you get it as a bonus when you install that (clone + requirements, plus the two Hunyuan checkpoints - the full drill is on the mesh generator page). There are no settings beyond picking the file; the dropdown lists images in ComfyUI's input folder, and you can upload straight from the widget.

The main thing to know is which output to wire where: image goes to the 3D generator or anything that wants plain RGB, mask goes to anything that wants a silhouette, and image_with_alpha is there when a downstream node genuinely needs the RGBA. Get the wiring wrong and you'll feed the generator a flattened image and never notice the transparency existed.