Guyanonymous' Prompt Lightbox

User guide · Version 1.0.0 · Released 3 November 2026

Introduction

Guyanonymous' Prompt Lightbox is an image viewer for AI-generated art. It works like a normal photo viewer: you can browse, zoom, pan, rotate and run a slideshow. It also has a panel docked on the right that reads the generation data hidden inside each image. That data includes the positive and negative prompts, the model, sampler, seed, steps, CFG scale, LoRAs and every other field the generator saved. *it's also very fast.

You can copy any part of that data with one click, or export it as JSON, as A1111-style text or as a Markdown summary. You can export one image or every image you have loaded at once.

Everything runs inside your browser. Images are read from your own disk and never uploaded anywhere. There is nothing to install: it is a single HTML file. When opening image collections and folders, the 'upload' the dialog refers to is local working space, not off the computer you're using.

What it reads

GeneratorWhere the data livesAutomatic1111 / Forge / reForgePNG parameters text chunk, or JPEG/WebP EXIF UserCommentComfyUIPNG prompt (API graph) and workflow (UI graph) chunksNovelAIPNG Comment (JSON) and Description chunksInvokeAIPNG invokeai_metadata, legacy sd-metadata and Dream chunksSwarmUIparameters JSON with sui_image_paramsFooocusparameters JSONMidjourneyXMP / PNG Description, including --ar, --v style flags and Job IDAnything elseEXIF ImageDescription, UserComment, XP fields, XMP dc:description, JPEG comments

Changes since the previous major version

Version 1.0.0 is the first major release. The previous version was the 0.1.0 pre-release, built earlier the same day. These are the changes between them:

• New name. The app is now called Guyanonymous' Prompt Lightbox. The name appears in the browser tab, the toolbar and the keyboard shortcuts window.

• Version label. A v1.0 badge sits next to the name in the toolbar and in the shortcuts window.

• Version in exports. Every JSON export now records exportedBy with the new name and a new appVersion field set to 1.0.0. This tells you which version produced a file.

• Standalone package. The app now ships as a self-contained HTML file in a zip, together with this guide in Markdown and HTML. You can open it from your own computer with no account and no connection.

The viewer, parser and export features are unchanged from 0.1.0. They are all described in the instructions below.

Getting started

1. Unzip the package anywhere on your computer. You will find: - Guyanonymous-Prompt-Lightbox.html (the app) - Prompt-Lightbox-Guide.md and Prompt-Lightbox-Guide.html (this guide)

2. Open the app. Double-click Guyanonymous-Prompt-Lightbox.html. It opens in your default browser. A current version of Chrome, Edge, Firefox or Safari works.

3. Look at the examples. Two example images load automatically: a lighthouse made with A1111 and a koi pond made with ComfyUI. Click between them in the filmstrip at the bottom and watch the right-hand panel change.

4. Load your own images. Drag image files or a whole folder onto the window. You can also click Open images in the toolbar, or press Ctrl+O (Cmd+O on a Mac).

5. Read the prompt. The panel shows which generator made the image, then the positive and negative prompts, then the settings.

6. Export. Click Export at the top of the panel and choose a format. Your browser saves the file to your usual downloads folder.

Tip: Images in the filmstrip with a small amber dot contain a prompt. Images without the dot have no generation data that the app could find.

Complete instructions

The screen

AreaWhat it doesToolbar (top)Opening files, navigation, zoom, rotation, display options, help and the panel toggleStage (centre)Shows the current image. Drag to pan, scroll to zoom.Status bar (under the stage)File name, pixel size, file size, format, rotation, and the pixel coordinate under your cursorFilmstrip (bottom)Thumbnails of every loaded image. Click one to view it.Prompt panel (right)The generation data for the current image, plus the Export menu

Opening images

• Open images: the picture-with-plus button. You can select several files at once.

• Open folder: the folder button. This loads every image inside the folder and its subfolders. The relative path shows in the panel as Folder path.

• Drag and drop: drop files or folders anywhere on the window. A dashed outline appears while you drag.

• Paste: copy an image in another program and press Ctrl+V / Cmd+V.

New images are added to the end of the filmstrip, sorted by name with numbers in natural order (img2 before img10). The first new image opens straight away.

Any format your browser can display will open, including PNG, JPEG, WebP, GIF, AVIF, BMP and SVG. Prompt data is read from PNG, JPEG and WebP files.

Moving between images

• Use the ‹ › toolbar buttons, the arrows that appear at the stage edges when you hover, or the ← → keys.

• Page Up / Page Down also move back and forward.

• Home jumps to the first image and End to the last.

• Navigation wraps around: going forward from the last image takes you to the first.

• The counter between the arrows shows your position, for example 3 / 12.

Zooming and panning

• Zoom at the cursor: scroll the mouse wheel or pinch on a trackpad or touchscreen. Zooming centres on the point under your cursor.

• Zoom buttons: the magnifier buttons, or the + and − keys. Each step is 25%.

• Fit to window: the four-corners button or 0. The whole image fits on screen. Small images are not enlarged past 100%.

• Actual size: the 1:1 button or 1. One image pixel equals one screen pixel.

• Switch between fit and 100%: double-click the image. When fitted, it zooms to 100% at the spot you clicked. Otherwise it returns to fit.

• Pan: drag the image, or hold Shift and press the arrow keys.

• The current zoom level shows between the zoom buttons. The range is 2% to 4000%.

• Above 250% the image switches to sharp, unsmoothed pixels so you can inspect fine detail.

• While the view is fitted, it re-fits automatically when you resize the window or the panel.

Rotating and flipping

• Rotate right: R or the clockwise-arrow button. Rotate left: Shift+R or the anticlockwise button. Each press turns 90°.

• Flip horizontally: H. Flip vertically: V.

• Active flips are highlighted in the toolbar. The status bar shows the rotation angle and flip symbols.

• These changes affect the view only. Your files are never modified. Each image opens in its original orientation.

Backgrounds, slideshow and fullscreen

• Background (B): cycles through Theme, Black, Grey, White and Checkerboard. The checkerboard makes transparent areas easy to see. Your choice is remembered next time.

• Slideshow (Space): advances every 3.5 seconds, and a Slideshow badge appears on the stage. Press Space or Esc to stop. Using the arrow keys also stops it.

• Fullscreen (F): fills the screen. Press F or Esc to leave.

Removing images

Press Delete or Backspace, or click the bin button, to take the current image out of the filmstrip. This only removes it from the viewer. The file on your disk is never deleted.

The prompt panel

Press I or click the panel button at the far right of the toolbar to show or hide the panel. Drag its left edge to make it wider or narrower. The width is remembered.

The panel shows these sections, top to bottom:

1. Generator badge. Which tool made the image, for example Automatic1111, Forge, ComfyUI or NovelAI. If no prompt is found it shows No prompt found and explains why. The two built-in samples also carry an Example image badge.

2. Positive prompt. The full prompt text, with a count of comma-separated tags and characters. LoRA and embedding tags such as <lora:name:0.6> are highlighted, as are weighted terms such as (fog:1.2) and the BREAK keyword. Copy puts the whole prompt on your clipboard.

3. Negative prompt. Shown the same way, in a quieter colour, with its own Copy button.

4. Parameters. Every setting the generator saved, such as steps, sampler, scheduler, CFG scale, seed, size, model, model hash, LoRAs, hires-fix settings and version. Click any row to copy that value, which is handy for reusing a seed.

5. File. Name, folder path, dimensions, aspect ratio, megapixels, file size in KB/MB and exact bytes, format and modified date. Camera details such as make, model, lens, exposure, aperture and ISO are listed when present.

6. Raw metadata. Every text field found in the file, labelled with where it came from (for example PNG tEXt · parameters or EXIF · UserComment) and its length. Click a field to expand it. JSON is pretty-printed. Copy field copies the original text. Very large fields, such as big ComfyUI workflows, are shortened on screen only; copying and exporting always use the full text.

How ComfyUI images are read

ComfyUI saves the whole node graph rather than a simple prompt. The app finds the sampler node (KSampler, SamplerCustom, or SamplerCustomAdvanced with a guider). It then follows its positive and negative connections back through the graph to the text-encode nodes. This works through LoRA loaders, ControlNet nodes, conditioning combiners and primitive or string nodes. It also collects the checkpoint or UNet model, every LoRA with its strength, the VAE, the latent size and the node count.

If an image contains only the UI workflow and not the API prompt, the app reads the text-encode nodes directly. It guesses which one is negative from node titles containing "neg", and adds a note in the parameters saying so.

How Midjourney images are read

The description text becomes the prompt. Trailing flags such as --ar 16:9 or --v 6.1 and the Job ID are split out into the parameters table.

Exporting

Click Export at the top of the panel to open the menu.

OptionFileContentsPrompt + all metadata<image>.prompt.jsonStructured record: file details, generator, positive and negative prompts, parameters, and every raw metadata fieldGeneration parameters<image>.parameters.txtA1111 format. Paste it into A1111/Forge's PNG Info tab or the prompt box to restore every setting. For A1111 images the original text is exported exactly as saved.Readable summary<image>.prompt.mdMarkdown with the prompts, a parameters table, file details and all raw fieldsCopy everythingclipboardThe same JSON as the first option, ready to pasteAll loaded imagesprompts-<N>-images.jsonOne JSON file with a record for every image in the filmstrip

Shortcut: press E to export the current image's JSON straight away.

<image> is the image's file name without its extension. For example, castle_0042.png exports as castle_0042.prompt.json.

JSON structure

{
  "exportedBy": "Guyanonymous' Prompt Lightbox",
  "appVersion": "1.0.0",
  "exportedAt": "2026-11-03T17:30:00.000Z",
  "file": {
    "name": "castle_0042.png",
    "path": "renders/castle_0042.png",
    "format": "PNG",
    "mimeType": "image/png",
    "bytes": 1482211,
    "width": 832,
    "height": 1216,
    "lastModified": "2026-11-02T21:14:09.000Z"
  },
  "generator": "ComfyUI",
  "prompt": {
    "positive": "…",
    "negative": "…"
  },
  "parameters": {
    "Seed": 604117752913,
    "Steps": 30,
    "LoRA": ["detail_xl.safetensors @ 0.7", "style_b.safetensors @ 0.4"]
  },
  "metadata": [
    { "source": "PNG tEXt", "key": "prompt", "value": { "…": "full parsed graph" } }
  ]
}

• When a setting appears more than once, as with several LoRAs, its value becomes a list.

• Raw metadata values that are valid JSON are stored as JSON objects, not strings.

• The All loaded images export wraps the records in { "exportedBy", "appVersion", "exportedAt", "count", "images": [ … ] }.

Keyboard shortcuts

Press ? at any time to see this list in the app.

ActionKeyPrevious / next image← → (or Page Up Page Down)First / last imageHome EndZoom in / out+ −Fit to window0Actual size (100%)1Rotate right / leftR / Shift+RFlip horizontal / verticalH / VPanShift + arrow keys, or dragCycle backgroundBStart / stop slideshowSpaceFullscreenFShow / hide prompt panelICopy positive promptCExport JSONEOpen imagesCtrl+O / Cmd+ORemove from listDelete / BackspaceKeyboard shortcuts?Close dialog / menu, stop slideshowEsc

Using it on a phone or tablet

On narrow screens the layout stacks:

1. Toolbar

2. Image

3. Filmstrip

4. Prompt panel, which you scroll down to read

Pinch to zoom, drag to pan, and use the on-screen arrows to move between images. Some desktop-only controls are hidden on small screens: the folder button, fullscreen and the panel toggle.

Troubleshooting

"No prompt found" on an image I know was generated. Most websites, Discord, social networks and messaging apps remove metadata when you upload an image. Saving a screenshot also loses it, as does re-saving the image in an editor. Use the original file straight from your generator's output folder.

The image shows "This image carries metadata, but none of it looks like a generation prompt." The file has text fields, such as camera EXIF or editing software tags, but none in a format the app recognises. Open Raw metadata to see everything that is there.

ComfyUI prompts appear joined with ---. The graph had no sampler node the app could trace, so it lists every text-encode node instead.

A file won't display. The app shows Couldn't display … in this browser. Formats such as HEIC or some TIFFs only open if your browser supports them. Convert the file to PNG or JPEG first.

Export does nothing. Your browser may be blocking downloads from local files. Allow downloads for the page, or use Copy everything and paste the result into a text file. When the app runs inside claude.ai, each save shows a confirmation prompt that you need to accept.

Fonts look different. The app loads its typefaces from Google Fonts. Offline, it falls back to your system fonts. Everything still works.

Privacy

• Images are read locally by your browser and are never uploaded.

• The app stores only two preferences in your browser: the panel width and the background choice.

• Nothing is sent anywhere.

Limitations

• Prompt data is read from PNG, JPEG and WebP only. GIF, AVIF, HEIC, TIFF and BMP open for viewing but are not scanned.

• NovelAI's hidden "stealth" metadata, which is stored in the alpha channel's pixels, is not decoded.

• C2PA / Content Credentials manifests (DALL·E, Adobe Firefly) are not decoded. Any XMP description they include is still read.

• Rotation and flipping affect the view only and cannot be saved back to the file.

• The loaded image list is not saved between sessions. Reopen your files or folder next time.

Full changelog

All notable changes since the app was first created.

1.0.0 (3 November 2026), first major release

Changed

• Renamed the app from Prompt Lightbox to Guyanonymous' Prompt Lightbox. The new name is used in the page title, toolbar and exports.

Added

• Version badge (v1.0) in the toolbar and the keyboard shortcuts window.

• appVersion field in all JSON exports.

• Standalone, offline-capable HTML build packaged in a zip.

• User guide in Markdown and HTML, with an introduction, changes since the previous major version, getting started, complete instructions and full changelog.

0.1.0 (3 November 2026), initial build

Viewer

• Open images through a file picker, a folder picker, drag and drop (including nested folders) and clipboard paste.

• Filmstrip of thumbnails with a marker on images that contain a prompt. Files sort in natural order.

• Previous / next and first / last navigation with wrap-around, plus edge arrows on the stage.

• Zoom at the cursor with the wheel or pinch, zoom buttons, fit to window, actual size and double-click to switch between them.

• Pixel-sharp rendering above 250%.

• Drag panning and keyboard panning.

• 90° rotation both ways, and horizontal and vertical flip.

• Five stage backgrounds, including a transparency checkerboard. The choice is remembered.

• Slideshow at 3.5 seconds per image, and fullscreen mode.

• Status bar with dimensions, file size, format, transform state and the live pixel coordinate under the cursor.

• Remove from list, which never deletes the file.

• Neighbour preloading for fast browsing.

• Resizable, collapsible prompt panel. The width is remembered.

• Keyboard shortcut overlay.

• Light and dark themes that follow your system setting.

• Layout for phones and tablets.

Metadata reading

• PNG tEXt, zTXt (compressed), iTXt (including compressed and XMP) and eXIf chunks.

• JPEG EXIF, XMP and comment segments.

• WebP EXIF and XMP chunks.

• EXIF fields including UserComment (ASCII and UTF-16 in either byte order), ImageDescription, Windows XP tags, camera and lens details.

• Interpreters for Automatic1111, Forge and reForge; ComfyUI (API graph tracing and UI-workflow fallback); NovelAI (v3 and v4 caption formats); InvokeAI (current and legacy); SwarmUI; Fooocus; Midjourney; and plain description text.

Prompt panel

• Generator badge.

• Positive and negative prompts with tag and character counts, highlighting for LoRAs, weights and BREAK, and one-click copy.

• Parameters table with click-to-copy rows.

• File and camera details.

• Expandable raw metadata fields with pretty-printed JSON and per-field copy.

Export

• Per-image JSON with full structured metadata.

• A1111-format parameters text.

• Markdown summary.

• Copy-all to the clipboard.

• Batch JSON for every loaded image.

• E shortcut for a quick JSON export.

Examples

• Two built-in example images, generated in the browser, with real embedded A1111 and ComfyUI metadata.