LoadImageWithAlpha

ComfyUI's built-in Load Image node converts everything to RGB - which is fine for a normal photo, and quietly destructive if the file you're loading is a transparent PNG where the alpha channel actually means something. That's the entire reason this node exists: it's the same upload widget you already know, but it converts to RGBA instead, so the alpha channel survives into the tensor instead of getting flattened away.

Why it matters for this pack

Avatar Graph's rigging pipeline works by segmenting a face into isolated pieces - an eye, a mouth - and building flat mesh geometry from their outlines. A lot of that pipeline runs on transparent cutouts, where "everything outside the alpha" is exactly the information the mesh-building steps need to define a shape's boundary. Load that through the stock RGB loader and you lose the one channel that mattered; the image comes back as a solid rectangle with whatever background color filled in, not a shaped cutout. LoadImageWithAlpha is what you reach for the moment transparency isn't decorative - it's data.

The input and output that matter

• image - the standard ComfyUI file-upload widget, same as the built-in Load Image node. Pick a file from your input directory or drag one in.

Output is a single IMAGE, RGBA this time rather than RGB, ready to feed into whatever in your workflow needs the alpha channel intact - typically one of the pack's mesh-generation or boundary-extraction steps.

How to install it

Via ComfyUI Manager: search "Avatar Graph" (avatar-graph-comfyui), install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/avatechai/avatar-graph-comfyui
cd avatar-graph-comfyui
pip install -r requirements.txt

The pack's hard requirement: bpy==3.6.0 is pinned in requirements.txt, and it only builds on Python 3.10.x. If your ComfyUI runs 3.11 or 3.12, this install fails before it gets anywhere near this node. Windows users get Avatech's own bundled Python 3.10 ComfyUI zip from the README; Mac/Linux users should build a dedicated conda environment (conda create --name comfyui python=3.10) instead of retrofitting an existing one. Expect a heavier install overall - segment-anything, mediapipe, and OpenCV all come along with it.

Common issues & troubleshooting

Your cutout came out as a solid rectangle instead of the shaped region. That's the symptom of using the stock Load Image node on a transparent PNG somewhere upstream - the alpha got flattened before it reached you. Swap in LoadImageWithAlpha at the point the file is actually loaded, not after.

Image looks the same either way. If your source file has no real alpha channel to begin with (a flattened JPEG, a PNG saved without transparency), this node has nothing extra to preserve - it'll just behave like a normal RGB loader in that case. The difference only shows up on genuinely transparent source images.

Not sure if a file actually has usable transparency. Open it outside ComfyUI first (any image viewer that renders transparency as a checkerboard) - if it looks like a normal opaque image there, there's no alpha data for this node to carry through regardless of which loader you use.