AI metadata

If you generate images with Stable Diffusion tools, monbooru reads the generation parameters they embed and makes them browsable and searchable. Extraction runs automatically at ingest for PNG, JPEG and WebP files; images without metadata are ingested without it.

What is read#

A1111 / Forge parameters are read from the PNG parameters text chunk, the JPEG EXIF UserComment field, or the WebP EXIF chunk. monbooru parses the positive and negative prompt plus the parameter line: steps, sampler, CFG scale, seed, model and LoRA hashes. The full parameter line is kept as well.

ComfyUI workflows are read from the PNG prompt chunk (the API format) with the workflow chunk as a fallback. monbooru follows the node graph to pull out the prompt, seed, steps, CFG, sampler, scheduler, checkpoint and LoRAs, and stores the complete workflow JSON.

An image can carry both at once (for example a ComfyUI render post-processed through Forge); ai:any matches either.

Where it shows#

The detail page renders a Generation data block: prompt, negative prompt, model, sampler, seed and the rest, with the full raw parameters or workflow available underneath.

At the top of the block sits the generation hash, a short fingerprint of the recipe (prompt, negative prompt, model, sampler, steps, CFG - deliberately not the seed). Every image generated from the same recipe shares the hash, so clicking it lists all your re-rolls of that prompt together.

Searching on it#

The related filters (ai:, prompt:, model:, sampler:, seed:, generated:) are documented in Searching. A couple of examples:

• ai:comfyui model:sdxl - ComfyUI images made with an SDXL checkpoint.
• prompt:catgirl seed:12345 - a specific render, found by its seed.

Re-extraction#

If you add files that were indexed before, or a monbooru update improves the parser, run Settings -> Maintenance -> Re-extract metadata to re-run extraction across the library. See Maintenance.