AI Image Metadata & Prompt Extractor

Read real embedded generation metadata from original AI images locally in your browser: NovelAI (standard + Stealth PNG Info + V4 multi-character), Stable Diffusion WebUI / Forge, and ComfyUI workflows.

100% Browser-Local · Zero Upload Free · No Membership
Important Format & Original File Notice: Upload original unedited PNG files (or metadata-preserving original JPEG / WebP). Screenshots, social-app compressed copies, and re-saved images strip embedded chunks and Stealth PNG bits, making recovery impossible.

Drop Original NovelAI / Stable Diffusion / ComfyUI Image

Supports PNG (incl. Stealth PNG Info), A1111/Forge, ComfyUI Workflow · ≤ 30 MB · 100% local

How to Use

1

1. Upload Original Unedited Image

Drop an original PNG export (or metadata-preserving JPEG/WebP). Compressed or screenshot copies strip metadata.

2

2. Automatic Format Detection

Scans PNG tEXt/zTXt/iTXt chunks, NovelAI Alpha/RGB Stealth LSB streams, ComfyUI node graphs, and A1111/Forge parameters.

3

3. Copy Prompts or Export Workflow

Copy positive/negative prompts and parameters, or download the raw ComfyUI workflow JSON.

Common Use Cases

  • Recovering NovelAI prompts, seeds, and V4 multi-character coordinates from original PNGs or Stealth PNG Info.
  • Extracting an embedded ComfyUI workflow.json from a rendered PNG to re-import into ComfyUI.
  • Checking exact model names, LoRA weights, CFG, and samplers from A1111 / Forge outputs.

Practical Example & Limits

Example: Dropping a ComfyUI PNG extracts Positive/Negative prompts, model name, sampler settings, and enables Workflow JSON download. Dropping a re-compressed screenshot reports honestly that no metadata exists.

Limits: Reads embedded metadata only (not AI image-to-text guessing). Social media uploads (Discord/X/WeChat), screenshots, and re-saved images wipe chunks and Stealth LSB pixels. Max 30 MB / 25M pixels, 100% browser-local.

Frequently Asked Questions

What is the difference between PNG chunks, Stealth PNG Info, and ComfyUI workflows?

A1111/Forge and standard NovelAI store text in PNG tEXt/iTXt chunks; NovelAI also hides gzipped JSON in pixel LSBs (Stealth PNG Info); ComfyUI embeds full prompt and workflow node graphs.

Why do compressed images show "No recognizable metadata detected"?

Platforms and image editors strip PNG chunks and re-encode pixels, permanently erasing embedded prompts. Original files are required.

How are complex ComfyUI custom nodes handled?

Known sampler/loader nodes are parsed while unknown custom nodes are flagged transparently with full raw Workflow JSON preserved — never fabricated.

Do you support NovelAI V4 multi-character prompts?

Yes, when present in v4_prompt.caption.char_captions, each character prompt and center position is listed cleanly.

Recommended Next Tools

All 13 Tools →