Load Image

At first glance DeepGen_LIMG looks redundant - ComfyUI already ships a LoadImage node, so why would this pack add its own? The answer is that this one does what the built-in doesn't: it reads XMP metadata back out of an image. Load a DeepGen-generated image here and you don't just get pixels - you get the description, dialogues, and assets strings that the generation wrote into the file, ready to wire back into your workflow.

Why you'd reach for it

The metadata is the point. DeepGen's pipeline writes human-readable sidecar info into generated files - a scene description, any dialogues, and an assets list - and this loader surfaces it. That turns the loader into a round-trip tool: load a previous render, grab its description, feed it into a fresh generator as a prompt, and the model has the full context it was created with. If you've built a workflow that chains generations, this is how you keep the chain's memory without re-typing prompts. For plain image loading with no metadata interest, ComfyUI's built-in node is fine; this one is for when the file's story matters.

The inputs

• image - a dropdown of image files in ComfyUI's input/ directory (png, jpg, jpeg, webp, bmp, gif), with an upload button. Same picker pattern as the built-in loader.
• force_height (0–8192, step 8, default 0) - set to a target height and the image is resized to it, keeping aspect ratio (width scales proportionally). 0 means "no resize, use the original." Handy for feeding the generators a consistent resolution.

The outputs

• image (IMAGE) - the pixels, as a normal tensor. Wire into any DeepGen generator or any standard image node.
• mask (MASK) - the alpha channel as a mask (inverted to ComfyUI convention; fully black if the image has no alpha).
• description, dialogues, assets (STRING) - the XMP metadata, parsed out of the file. These are the loaders' special sauce; feed them to a prompt input or log them. If a file has no metadata, they come back empty strings, which is fine.

Install

Same pack as the rest of the DeepGen nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/deepiksdev/ComfyUI-DeepGen-API.git
cd ComfyUI-DeepGen-API
pip install -r requirements.txt

Restart ComfyUI. Unlike the generator nodes, this one needs no API key - loading is local, pure file reading.

Gotchas

The metadata is only as good as the file: images generated by other tools, or stripped of XMP, will return empty strings - don't expect the pack's magic fields on every PNG. And force_height upsizing a small image won't conjure detail, it just resizes; keep expectations honest. If you load a video with the pack instead, that's DeepGen_LVID, which does the same metadata read for video files.