JsonParse (Yogurt Nodes)

ComfyUI hands you strings everywhere - API responses, saved workflow metadata, an LLM node's "structured output" that came back as text. And a string of JSON is just a string until you parse it. JsonParse is the conversion: feed it the text, get back an actual object (dicts, lists, numbers) that other JSON nodes - GetPath, Merge, Stringify - can work with.

It's a logic node in the YogurtNodes pack (yogurt7771/ComfyUI-YogurtNodes), and the usual entry point to the pack's JSON toolkit. If you're doing anything where a model or a web service hands you JSON-as-text, this is step one.

How it works

Straightforward: it runs Python's json.loads on the input and returns the resulting object as json_object. The interesting knob is strict (default on). Strict mode is the spec-compliant json.loads - it will reject single-quoted strings, trailing commas, and comments. Turn strict off and the node first swaps single quotes for double quotes before parsing, which rescues a lot of sloppy-but-common JSON emitted by LLMs. That's the mode to reach for when your model "forgot" to use proper double quotes.

If the input genuinely isn't valid JSON, the node raises a clear Invalid JSON error rather than silently returning garbage - which is the right behavior, because garbage would otherwise flow through your whole graph.

Inputs that matter

• json_string - the text to parse. Multiline widget, so paste freely.
• strict - on by default. Turn off only when you're dealing with single-quoted output from a model; it's a heuristic, not a license to feed it anything.

Output: json_object.

Install

Pack-wide routine - ComfyUI Manager (search "YogurtNodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes
pip install -r requirements.txt

Restart, find it under Yogurt Nodes / Logic. No dependencies beyond the standard library.

Troubleshooting

• "Invalid JSON" and I copied it from a model. Classic LLM output problems: single quotes, trailing commas, markdown code fences around the JSON. Turn strict off for the quote issue; strip backticks before parsing for the fence issue. If there's prose wrapped around the JSON, you need a node that extracts the JSON first - this one expects clean input.
• "It parsed but everything is one giant string." You got a JSON string containing a JSON string - double-encoded. Run it through Parse again (or fix whatever double-encoded it upstream).
• Leading BOM / whitespace. Rare, but some APIs prepend a byte-order mark that makes json.loads complain. If you see a cryptic error on input that looks fine, that's the suspect.
• Type expectations. A JSON array parses to a list, an object to a dict, true/false/null to Python booleans/None. If a downstream node expects a dict and you fed it an array, that's a shape mismatch, not a parse failure.

Rule of thumb: Parse at the boundary - right where text enters your graph - and everything downstream gets to think in objects instead of strings. Pair it with JsonValidate when you want to confirm the shape before trusting it.