JSON Extractor

ComfyUI workflows are JSON. Generated images carry their workflows as embedded metadata, which is JSON inside the PNG. API responses from image models are JSON. And yet, in a graph editor, JSON has historically meant "paste a blob into a text node and squint." JSON Extractor is the node that ends that: you feed it a JSON string and a dotted key path, and it walks the structure and returns the value - in four formats at once.

The README's own pitch is parsing metadata, and that's the workflow that sells it. You load an image that has a workflow embedded, extract the JSON (the pack's Load Image With Metadata does this natively), and then workflow.settings.cfg or nodes.3.seed becomes a live value you can wire straight into a KSampler. One loaded image, and your graph re-derives the settings that made it.

How it works

It parses the JSON, then walks the key_path dot-notation: workflow.settings.cfg means dig into workflow, then settings, then grab cfg. The path language is small but genuinely useful:

• Object keys - prompt.something
• Array indices - nodes.0.seed reaches into the first element of a list
• Nested paths - data.items.2.name
• Nested JSON strings - if a field's value is itself a JSON string and more of the path remains, it parses it on the fly. This is the trick that makes embedded-workflow metadata work, because that metadata is often a JSON string inside the JSON

Once it finds the value, it converts it to all four types: string, float, int (handles "7.5" → 7 via float), and boolean. Missing paths and failed conversions return None on that output rather than erroring, which is friendly for graphs - the node won't crash a run just because a key wasn't there.

Inputs and outputs

• json_string - the JSON text, wired in (forceInput), not typed
• key_path - the dotted path, e.g. workflow.settings.cfg or nodes.0.seed

Four outputs, one per type: string, float, int, boolean. Take whichever you need and ignore the rest - they're all the same value, just coerced. Wiring the int output into a seed or step input and the string output into a prompt port is the intended pattern.

Installing it

One node in the XJNodes pack. ComfyUI Manager → search "ComfyUI-XJNodes" → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/alexjx/ComfyUI-XJNodes
# restart ComfyUI

It uses Python's stdlib json (plus a small compatibility shim for tolerant parsing), so no extra packages. Category: XJNodes/text.

Where it gets fiddly

The failure mode to learn: a path that doesn't exist doesn't raise, it silently returns None. So a typo like workflow.setting.cfg looks fine until the value doesn't flow. If a node downstream suddenly receives None, check the path before suspecting the wiring. Also, _to_bool uses Python truthiness - the string "false" converts to True because it's a non-empty string, which will absolutely bite you if you're extracting booleans from JSON. For numbers and strings it's a workhorse; for booleans, verify the output once. If you're feeding values from an API response rather than embedded workflow metadata, the same rules apply - and honestly, this node is the rare utility in a personal-use pack that deserves a permanent spot in your toolset.