AI Metadata Viewer Now Reveals Hidden Prompts Inside Any AI Image

By vramkickedin | | 2 min read
An elegant magnifying glass crafted from translucent matte acrylic tiny floating luminous icons.

A small, privacy-first web tool called AI Metadata Viewer now gives anyone a quick way to read all the hidden creation data tucked inside AI-generated images. You simply drag a PNG or JPEG from Stable Diffusion, ComfyUI, NovelAI, or Midjourney onto the page and it instantly reveals the prompt, model, sampler, seed, LoRAs, and even full ComfyUI workflows. Everything is parsed right inside your browser—no upload, no server, no account needed.

Developer GChenSi-2 built this because most existing metadata viewers dump raw, hard-to-read text strings at you. They created a client-side parser that transforms that chaos into clearly labeled fields you can actually use, and then wrapped it all in a clean interface that works in English, Chinese, and Japanese. The tool also quietly saves your last 20 lookups locally, without storing any image files.

One-click workflow extraction from any image

Key Features
  • Extracts metadata from PNG chunks locally.
  • Surfaces model, sampler, prompts, and LoRAs.
  • One-click download of ComfyUI workflow JSON.
  • Keeps a local-only history of last 20 entries.
  • Supports English, Chinese, and Japanese languages.
  • Accepts drag-and-drop or clipboard paste.
  • Never sends your image to any server.

Artists and prompt engineers who frequently switch between Stable Diffusion, ComfyUI, and other platforms can instantly inspect how an image was made or recover a lost workflow. Privacy-conscious users get a safe way to examine sensitive projects because nothing ever leaves their computer. Hobbyists learning from public images can simply drag any compatible file to understand the exact parameters used.

Built for privacy and speed

The entire parsing engine is written in pure JavaScript and requires zero backend services—the site deploys on Vercel’s free Hobby tier and runs offline once the page loads. Unit tests cover all four parsers using synthesized PNG fixtures, so no binary blobs clutter the open-source repository. The local history caps at 20 entries and doesn’t store images, and the developer welcomes community contributions, especially sample files that the current parsers handle poorly.

“Never uploads your image — the parser is pure JS, runs offline once the page is loaded” — Source: GitHub

Project Links