JSON process

Feed it a JSON string, give it up to four paths, get back up to four extracted values. That's json_process (the class name is lowercase, don't go hunting for a capitalized version). It's the natural companion to ImageMetadataReader: that node hands you a big JSON blob of PNG metadata, and this one digs the specific bits out - the width, the height, the A1111 parameters line, the SillyTavern character card.

The inputs:

• json_data - the JSON string. It's forceInput, so you wire it from a string output (like ImageMetadataReader's metadata_json); you can't just type a blob into it.
• query_1 … query_4 - the paths to extract, with a proper little query language: key1.key2[0].key3[2] for nested keys and array indexing. The author's tooltip spells it out: "支持key1.key2[0].key3[2]这种查询语法" - i.e. dotted keys, square-bracket indices, mixed freely. Defaults are width, height, and png_text.parameters, which line up exactly with what ImageMetadataReader emits.

The outputs: value_1 … value_4, strings. A scalar becomes its string form; a nested dict or list becomes pretty-printed JSON (indent 2), which is a thoughtful touch - you can grab a whole object without losing structure. A missing or malformed query returns a friendly 错误: (error) message instead of killing the run.

Mechanically it's a small regex-driven path walker: split the path on dots (but not dots inside a number) and brackets, then descend. It handles the "wrong type" cases - indexing a non-list, keying a non-dict - and reports which key failed, which is exactly the kind of error message that makes JSON digging bearable.

Where it earns its keep: any workflow that consumes metadata. Pull png_text.parameters off a found PNG and feed it to a text model; pull png_text.chara off a SillyTavern image and route it to TextEncoderDecoder for the base64 card; grab a seed or width out of a workflow JSON. If your JSON comes from somewhere else entirely (an API response in a string), it works there too - it doesn't care where the string came from.

One honest note: the default query paths are clearly tuned for ImageMetadataReader output, so if you're pointing it at a different JSON shape, remember it's four independent queries - empty query means empty output, and each query is optional to fill.

Install

ComfyUI Manager → search comfyui-spawner-nodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/spawner1145/comfyui-spawner-nodes

Deps: piexif, pypng, xmltodict - light pure-Python, no models. README is a stub, UI labels are Chinese, author is spawner1145 (of a Wan2.1 SD extension). Works, thinly documented.

Troubleshooting

• "无效的 JSON 字符串" - the input isn't valid JSON. Usually you wired the wrong string, or the source emitted an error object (ImageMetadataReader returns error JSON for unreadable files - read that first).
• "错误: 未找到键 'parameters'" - the path doesn't exist in this particular JSON. Check the actual shape with a JSON viewer, or query png_text first to see what keys exist.
• All four outputs empty - you left queries blank. An empty query returns an empty string by design.