Load Image Prompt Metadata

"Workflow included" is the community norm for a reason: every ComfyUI-generated PNG carries the whole pipeline as embedded metadata, and dragging the image back into ComfyUI rebuilds the graph. HYBS_LoadImagePromptMetadata rides that same mechanism in the other direction - it loads an image and pulls the positive and negative prompts out of its embedded metadata, so you can reuse them in a new workflow without retyping.

The obvious use is img2img/i2i: someone posts a great image, you drop it in, and instead of eyeballing a screenshot to transcribe the prompt you get it as a clean text output you can wire straight into a CLIP Text Encode node. The less obvious use is your own archive - when you re-edit an old generation, the exact prompt comes along for the ride.

How it works

image is a combo dropdown of files in ComfyUI's input folder, just like Load Image. The node reads the PNG's metadata, looks for the standard workflow and prompt keys (plus a few variants), parses the JSON, and walks the node list looking for the node IDs you name. When it finds a match it takes the first string it can find among that node's widget values - which for a CLIP Text Encode node is your prompt text.

Two string inputs tell it what to find:

• positive_node_id - node ID to read as the positive prompt
• negative_node_id - node ID to read as the negative prompt

Both support subgraph IDs like 82:78 (parent:child), which matters for workflows that use subgraphs. Outputs are IMAGE (the loaded image, so you don't need a separate Load Image), plus positive and negative strings.

Getting the node IDs

Here's the part that confuses people. You need the node IDs from the source workflow's metadata, not from your current graph. The fastest way: open the source image in ComfyUI (drag it onto the canvas or into the window), let the workflow reconstruct, then look at the IDs on the CLIP Text Encode nodes - they're the big numbers in the node title bar. Type those into this node's inputs. The author ships an example in the repo (workflow/LoadImagePromptMetadata.json + a sample PNG) if you want to see the pattern.

If the source image has no metadata, or the ID you gave doesn't resolve to prompt text, the node raises an error rather than silently returning garbage - so a blank result usually means "that ID isn't a text node" or "this image came from A1111/Forge", which use a different metadata format this node isn't built to parse.

Installation

Standard pack install - no model downloads:

cd path/to/ComfyUI/custom_nodes
git clone https://github.com/hybskgks28275/ComfyUI-hybs-nodes.git
pip install -r ComfyUI-hybs-nodes/requirements.txt

Or "ComfyUI-hybs-nodes" in ComfyUI Manager, then restart. The only pip dependency is toml, which this node doesn't touch.

One honest caveat: if you only need the prompt, ComfyUI already lets you drag an image in and copy the prompt from the workflow tab. This node earns its place when you're automating - wiring the recovered prompt straight into an encode node, or rebuilding a pipeline where the prompt must live as graph data. That's a real niche, and if you're in it, this is a tidy little node.