NAI Meta Loader

Every NovelAI image you've ever generated is carrying its entire recipe around inside the file: model, sampling settings, the positive and negative prompts, even the per-character breakdown. NAI Meta Loader is the node that opens that recipe up and lets you see it - then hand it to the rest of the workflow piece by piece. Think of it as the "read" companion to NAI Meta Apply's "write": Meta Apply re-applies everything at once with checkboxes; Meta Loader shows you the whole card and lets you send individual values where you want them.

How it works

Drop an image onto the node (or select it), and the pack's backend reads the embedded metadata and lays it out in the same card structure the companion 4A Prompt Manager uses for its metadata cards: the model, the generation settings, the positive and negative prompts, and one card per character. From there you get per-item actions - send just the model, just the settings, just a prompt, or append a single character - plus a combined prompt action that replaces your complete prompt and character set in one move.

Like NAI Meta Apply, this is a UI-only node. It has a single input, imported_json (a multiline string), which is where the frontend stuffs the parsed metadata when you drop an image. There are no outputs: everything it does happens through the pack's interface and its server-side routes, not through the graph. That means it's also free - reading metadata is local, no token, no Anlas.

When you'd actually reach for it

Two honest use cases, and they're both about not retyping. First, you generated something on NovelAI's website and want to reproduce it in ComfyUI - Loader lets you inspect exactly what was used (including settings that are easy to mis-remember, like the exact sampler) and push it into your Scheduler and Sampler without transcription errors. Second, you have a character card worth keeping - the per-character view plus the append-character action is the fastest way to pull a single character out of an old image without clobbering the rest of your prompt.

The same metadata caveat as everywhere in this ecosystem

The metadata is in the file, not in your memory of the file. If the image was re-encoded - reposted to Reddit, previewed through Discord, converted to JPEG - the chunks are gone and the Loader will still show you something for the positive/negative text, but the model and generation settings won't be interpreted as NovelAI settings. So the golden rule applies here too: keep the original PNGs. A recompressed copy is a picture, not a workflow.

Install

Part of the ComfyUI-4A-NovelAI pack: ComfyUI Manager (search 4A NovelAI) or

cd ComfyUI/custom_nodes
git clone https://github.com/tsukino4a/ComfyUI-4A-NovelAI.git ComfyUI-4A-NovelAI
cd ComfyUI-4A-NovelAI
python install.py

Restart ComfyUI afterward. No model files, no heavy downloads - msgpack is the only real dependency. Pair it with NAI Meta Apply when you want the whole recipe back, and you'll stop hand-copying prompts faster than you'd think.