Append Input To JSON Array File

If you've ever run a ComfyUI batch, waited for the whole thing, and then realized you never captured the individual results anywhere - this is the node for that. Append Input To JSON Array File grows a JSON array on disk one entry per execution. Every run adds a new item to the list, so you end up with a structured record of everything the batch produced, instead of a stack of images and a vague memory.

It's part of JasonHoku's Lite Text to File Tools pack, and it's the natural companion to a batch that emits a different string each pass - say, an LLM captioning node labeling each image, or a filename generator. Feed it the string, it appends to the array. Nothing smarter, nothing sneakier.

How it works

Two inputs: text_input (the string to store) and file_prefix (the file path, relative to ComfyUI/output/). The default is text/lite-text.json, so out of the box you're writing to ComfyUI/output/text/lite-text.json.

On each run it reads the existing file, makes sure the data is a list, appends your new string, and writes it back with nice two-space indentation and UTF-8 encoding. Three edge cases are handled for you:

• File doesn't exist yet → starts with an empty array and appends the first item.
• File is corrupted / not valid JSON → discards the garbage and starts fresh (your bad data is gone, but the workflow keeps running).
• File is a JSON object instead of an array → wraps the existing object into a one-item list rather than erroring.

The status output reports what happened, including a running count: ✓ Appended to ... (total items: 7). It's an output node, so you can hook that status into a Show Text node to watch the array grow while a long batch runs.

What you'd use it for

• Dataset building. Run a batch, pipe each caption or prompt into this node, and you've assembled a JSONL-ish array you can train or review later.
• Prompt history. Log every positive prompt you generated with, then query it later with the pack's Load Text From JSON Array By Index node.
• LLM/VL output collection. This pack's README literally calls out batching LLM vision outputs - the array is a tidy place to accumulate them run by run.

Installing it

It's one of nine nodes in the Lite Text to File Tools pack, so the install is shared:

cd ComfyUI/custom_nodes/
git clone https://github.com/JasonHoku/comfyui-lite-text-to-file-tools
# restart ComfyUI

Or search "Lite Text to File Tools" in ComfyUI Manager. No extra Python packages, no models - just the standard library.

Gotchas

• Errors are returned as a status string, not raised, so a bad file_prefix won't stop the batch - you'll just get ✗ Error: ... in the status.
• There's no cap and no dedupe. Run the same prompt ten times, you get ten identical entries. That's on you.
• Want a fresh array for a new run? That's what the pack's Clear JSON Array File node is for - wire it first and this node starts from zero.