RC Save Image (No Metadata)

ComfyUI has a lovely habit: every PNG it saves carries your full prompt and the entire workflow JSON embedded in the metadata. "Workflow included" became the community norm for a reason - drag an image back into ComfyUI and the whole graph reconstructs. But sometimes that's exactly what you don't want. Post that image to social media and you're handing out your carefully tuned prompt, your LoRA weights, your secret sauce, for free. That's the problem RC Save Image (No Metadata) exists to solve: it's a drop-in replacement for the standard Save Image node that writes the file with the metadata stripped out.

It's the first node in the RC Image Compositor pack I'd grab even if I never touched a blend mode. The pack is a 30+ node Photoshop-style toolkit from kj863257, but this one utility earns its keep on its own.

How it works

The implementation is refreshingly honest. The node subclasses ComfyUI's own SaveImage and overrides the save routine to pass pnginfo=None when writing the PNG. That's the whole trick - no API, no exif-stripping library, no black magic. It saves to the same output folder with the same counter-based naming you're used to, so it behaves like the built-in node in every way except what gets written into the file.

The inputs that matter

Only two, and neither will surprise you:

• images - the IMAGE tensor to save.
• filename_prefix - default ComfyUI. Supports the usual formatting, so you can do %date:yyyy-MM-dd% or pull in values from other nodes.

It also passes the images through as an output, so you can keep the save in the middle of a graph and keep feeding the data onward instead of treating it as a dead end.

How to install it

The pack is one install, and this node comes with it:

cd ComfyUI/custom_nodes
git clone https://github.com/kj863257/ComfyUI_RC_Image_Compositor
cd ComfyUI_RC_Image_Compositor
pip install -r requirements.txt

Then restart ComfyUI. If you use ComfyUI Manager (and you should), just search for "RC Image Compositor" under Install Custom Nodes. Dependencies are light - just pillow, numpy, and opencv-python, which you almost certainly already have. No models to download, no API keys.

Common issues

The main gotcha is a conceptual one: this strips metadata at save time. If your graph loaded an image that already has embedded workflow data and you save that tensor through this node, the fresh PNG won't carry the prompt or workflow - because the node rebuilds the PNG from the tensor and writes no text chunks. If that's not the behavior you wanted, the pack's sibling node, RC Save Image (Transfer Metadata), lets you pull metadata back in from a source file on purpose. And if you just want a quick inline preview that doesn't pollute your output folder, RC Preview Image (No Metadata) is the same idea but writes to temp.

The other thing people bump into: it's not an encryption tool. Metadata being absent doesn't mean the image is untraceable - but for keeping your prompt and graph private when you share a render, it does the job.