JSON To Image Generation

If you've ever had a spreadsheet of prompts that each need to become an image, you know the manual way: paste, generate, save, repeat. YG JSON Prompt Loader ("JSON To Image Generation") automates the whole loop by reading a JSON or JSONL file where every entry is one job - a prompt plus an output_path saying where that image should end up. Each queue run hands you the next entry's prompt, so you can set the queue running and walk away.

How it works

Drop a .json or .jsonl file into ComfyUI's input folder, and it appears in the json_file dropdown. The file can be a plain JSON array of objects, or JSONL (one object per line) - the node sniffs which it is. Each object should look like:

{"prompt": "a devotional poster, warm tones", "output_path": "images/devotional/001.jpg"}

output_path is optional; if missing, it falls back to <id>.png or image.png. On each queue run the node emits the next entry and advances an index (the IS_CHANGED NaN trick keeps it re-executing every run). When the last entry is served, is_last flips true. The file's folder structure lives inside each output_path, which is the whole trick of the paired downloader node.

The natural wiring the author had in mind:

YG JSON Prompt Loader → [any image generator] → YG JSON Image Downloader

The inputs and outputs that matter

• json_file - the dropdown of JSON files in your input folder. Not seeing yours? Drop it in and restart, or use the override below.
• json_path_override - paste a full or relative path to a JSON file and it bypasses the dropdown entirely. Useful when your file lives elsewhere or changes often.
• job_id - must match the job_id on your YG JSON Image Downloader. Get this wrong and nothing breaks loudly, it just silently mismatches jobs.

Outputs: prompt (feed it to your text encoder or generator), output_path (feed it to the downloader), plus job_id, current_index, total_count, and is_last for batch bookkeeping.

Installing it

Part of comfyui_yg_nodes (MIT, "YG Custom Nodes" v1.0.4 by Mallikarjuna Rao Koliparthi). No extra Python dependencies. Install via ComfyUI Manager (search comfyui_yg_nodes) or:

cd ComfyUI/custom_nodes
git clone https://github.com/mallikarjunaraokoliparthi/comfyui_yg_nodes.git

Restart ComfyUI; it's under YG Custom Nodes.

Where people get burned

The state is in-memory and keyed by job_id: restart ComfyUI and the loader starts back at entry zero, so a 500-line JSON that was halfway through will restart from the beginning - and rerun the first 250 images unless you trim the file or change job_id. Also, "No entries found" usually means your JSON is neither an array nor valid JSONL - check for a stray trailing comma or a JSON object without brackets. And the loader itself never re-reads a changed file mid-job for the same job_id; edit the file, restart (or use a fresh job_id). For prompt-driven batch generation at scale - exactly the "Nano Banana Pro poster line" use case it was built for - this is a solid, dependency-free way to get there.