ImageOutputToComfyNodes

ImageOutputToComfyNodes is a format bridge. It takes an IMAGE tensor and returns an IMAGE tensor, but in between it converts the data back into the exact float-normalized form ComfyUI's own image nodes expect - so an image that came out of the pack's canvas side (PIL-based, 0–255) can be fed safely into the rest of your graph. One input, one output, no settings.

What it actually does

The source is three lines: it takes the first frame, converts it to a float32 numpy array, divides by 255 to normalize into 0–1, and wraps it back into a torch tensor with a batch dimension. That's it. The value here is purely about keeping data formats consistent. ComfyUI's IMAGE type is standardized on 0–1 float tensors, and the pack's canvas/outpaint tools produce PIL images that live in 0–255 land. When you're moving between those two worlds, this node is the adapter that makes the transfer clean instead of silently sending garbage ranges into a sampler.

When you need it

The pack's canvas workflows (CanvasOutpaint.json, the CanvasTool setup) route images through Python/Pillow and the Flask canvas server. When a result comes back from that side and you want to continue processing in regular ComfyUI nodes - or send it to a generation node as an input - running it through ImageOutputToComfyNodes first is the safe move. It's also a useful reference if you've ever wondered what the minimal correct "IMAGE → IMAGE" conversion looks like: normalize to 0–1, add the batch dim.

Setup

Part of taabata/LCM_Inpaint-Outpaint_Comfy:

cd ComfyUI/custom_nodes
git clone https://github.com/taabata/LCM_Inpaint-Outpaint_Comfy
cd LCM_Inpaint-Outpaint_Comfy
pip install -r requirements.txt

Or ComfyUI Manager, search LCM_Inpaint-Outpaint_Comfy. No models required.

Honest take

For most people this node will never appear in a workflow. It's a helper that exists because the pack's canvas tooling crosses a format boundary that normal ComfyUI nodes never cross - when everything stays inside ComfyUI, you never need it. But if you're following one of the pack's canvas tutorials and images look wrong (washed out, clipped, or a node rejects the input), the fix is usually that something crossed that 0–255 / 0–1 boundary without this node in between. Cheap insurance for a very specific problem.