Add Info

Here's the dirty secret about ComfyUI's built-in "the workflow is saved in the PNG" feature: it doesn't save much. The embedded prompt only records the values you typed into widgets. Node outputs? Gone. Values that arrived over wires from other nodes? Gone. So when you later load that image and wonder "what did the upscaler actually output, or what seed did that randomize node pick," the answer is: the file never knew.

Add Info fixes that at the moment of saving, and it's the node that makes the rest of the cg-prompt-info pack worth having. Drop it just before your Save Image node, and the PNG you write carries inputs and outputs for every node that already ran in the queue.

How it works

Add Info is a pass-through: an IMAGE goes in, the same image comes out. What it does on the side is the interesting part. When it executes, it reaches into ComfyUI's running executor and grabs the actual outputs of every node that has executed so far in this run - not just what's in the static workflow - and writes them into the PNG metadata that Save Image embeds.

A few honest limits on what lands in the file. Only values that are Python str, int, or float are stored properly. Big objects - checkpoints, models, latents, full image tensors - aren't, because dumping those would bloat the file and they're not comparable text anyway. Instead you get the first 30 characters of their string representation, which is usually just enough to tell you the thing existed and roughly what it was. The author says it plainly in the README: that first-30-chars trick "is normally enough to tell you what the thing was."

It also only records nodes that have already run. Put Add Info before a node in the graph and that node's outputs won't be in the file, because it hasn't executed yet. Put it just before you save and you capture everything upstream.

The inputs and outputs

• image (IMAGE) - the image you're about to save. Route your output here, then from here into Save Image.
• image out (IMAGE) - unchanged pass-through of whatever came in.

That's it for visible ports. The node also takes hidden prompt and extra inputs that ComfyUI injects automatically because the node declares them - you never wire them, they're how it gets at the metadata to modify.

Install

Same as the rest of the pack: zero dependencies, zero model files. ComfyUI Manager, search "Prompt Info", install and restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/chrisgoringe/cg-prompt-info

Restart, and it'll be under the prompt_info category in the node menu.

Common issues

People usually hit the "why is this node even here" wall first, because visually it does nothing. That's expected - it's a metadata side effect, not a transformation. If you're looking at an image you saved and the info output is thin or empty, ask whether Add Info was in the chain when that image was generated. You can't enrich an image after the fact; the metadata was decided at save time.

Also don't mistake the 30-character truncation for data loss you can fix. It's the design. If you genuinely need the full value of a node output, that belongs in the workflow as a string, not smuggled through metadata.

The one real habit to build: make Add Info a permanent part of your "save" branch. It costs one node and a few KB per file, and it turns every PNG you archive into a searchable record of exactly what ran. You'll thank yourself in a month when you're digging through a folder of old renders wondering which seed that batch actually used.