JSON to Multi-Line Summary

Every serious video workflow eventually dumps a JSON blob on you - inference params, a sampler's config, a VLM's output - and you stare at it and think "I just want to know what seed it used." JSON to Multi-Line Summary is that: it takes a dict or a raw JSON string, drills down to whichever part of it you care about, and emits a tidy multi-line string of just the keys you asked for.

It's the more flexible sibling of the pack's JSON Params Extractor. That node has a fixed key list; this one lets you name your own keys and your own path into the structure.

How it works

Three knobs on the right side, and they're the whole show:

• sub_path - where in the JSON to look. Defaults to inference_params, which is where Wan-style pipelines stash the good stuff. But here's the upgrade over a flat reader: it supports bracket indexing, so loras[0] grabs the first LoRA entry and foo.bar[2][1] walks a nested list. Empty means "use the whole object." If you land on a list, it quietly falls back to the first item - the common loras case.
• keys - comma-separated list of keys to extract, defaulting to the usual video suspects: height,width,num_frames,fps,guidance_scale,seed,steps and a heap more. Missing keys print as N/A instead of crashing.
• separator - what goes between key and value. Default ": " gives you seed: 1234. Change it if your downstream logger wants a tab or nothing.

The json_in socket accepts either a JSON or a STRING type, so you can wire it directly to a node that outputs structured JSON, or feed it pasted text. It tolerates both, which removes a whole class of type-mismatch friction.

The output

One STRING output, named plainly STRING. Wire it into any Preview Text, text-logging, or "show text" node and you get:

height: 704
width: 1280
num_frames: 121
seed: 123456789

Exactly the shape you want in a log line or an easy node's display.

Install

It's part of TrentNodes, the pack this whole family ships in:

cd ComfyUI/custom_nodes
git clone https://github.com/TrentHunter82/TrentNodes.git
cd TrentNodes
pip install -r requirements.txt

Or search "Trent Nodes" in ComfyUI Manager and restart. It shows up under Trent/Utilities. No models, no weights, no GPU - it's a pure formatting node.

Gotchas

Three things to know. First, invalid JSON doesn't crash the workflow - it returns a [JSONSummary] ERROR: string that you'll see in your preview, which is friendlier than a hard fail but easy to mistake for a real value if you're piping it into a logger. Second, the default keys list is tuned for Wan-style inference params; if your JSON uses different names, replace the whole list or everything prints N/A. Third, remember sub_path is relative - loras[0] means "under the top-level loras key," not "under inference_params.loras," so you may need inference_params.loras[0] to descend twice. Get that wrong and you get N/A lines, not an error - easy to chase.

Boring in the best way. When you need to know what a run actually did, this is the fastest wire from blob to eyeballs.