JsonFlatten (Yogurt Nodes)

Nested JSON is great for storage and awful for working with in a graph. JsonFlatten takes a deeply-nested object and collapses it into flat, dot-separated keys - {"settings": {"model": "xl", "steps": 20}} becomes {"settings.model": "xl", "settings.steps": 20} - which is a shape you can iterate, filter, and reason about one row at a time.

It's a logic node in the YogurtNodes pack (yogurt7771/ComfyUI-YogurtNodes), and it pairs with JsonUnflatten to make a round trip: flatten for processing, unflatten when you need the nested structure back. This is the classic "convert JSON to key-value rows" trick, and it's especially useful when you're handing structured data to an LLM node that wants a linear list, or building a table-like intermediate.

How it works

The node walks the object recursively. Every nested key gets joined with separator (default .), so depth becomes dots: a.b.c. Arrays get included by default (include_arrays, default on) as indexed keys - items.0, items.1 - so a list becomes a predictable sequence of flat keys rather than a lump. Turn include_arrays off and arrays get skipped entirely, which you want when array contents are irrelevant to what you're building.

Output is flattened_json: a flat dict where every value is a leaf - a string, number, or boolean. That's what makes it play well with other nodes: flat keys are trivial to filter with a regex, and flat values are trivial to sort or count.

Inputs that matter

• json_data - the nested object.
• separator - default .. If your keys themselves contain dots (say, filenames), pick something else like > or / to avoid ambiguous output. The unflatten side must use the same separator, or the round trip breaks.
• include_arrays - default on. Off if you want to ignore array contents.

Output: flattened_json.

Install

Same pack-wide install - 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, look under Yogurt Nodes / Logic. No dependencies beyond the standard library.

Troubleshooting

• "My flat keys look wrong." If a nested key already contains your separator (a . in a key name), the flatten is ambiguous. Change separator to a character that can't appear in your keys.
• "The round trip changed my data." Flatten → Unflatten is lossless if the separator matches and you keep include_arrays consistent on both sides. Mismatched settings are the #1 source of "where did my structure go."
• Ordering. The flattened output follows the traversal order, which for dicts means insertion order. Don't rely on it being sorted - sort downstream if you need deterministic order.

The practical move: flatten a config, show it as text, let an LLM or a human eyeball the whole config in one flat list, then unflatten the edited version. Much easier than asking a model to edit a deeply nested blob and hoping it doesn't drop a brace.