ImageLabs

1. Getting started

1. Launch ImageLabs.
2. Open an image: Open image (top-right, Ctrl+O), drag & drop a file onto the window, or Ctrl+V to paste a screenshot or copied image. Supported: PNG, JPEG, BMP, GIF, TIFF, WebP, ICO, HEIC, TGA.
3. Pick a task from the left rail (or use a Quick action), adjust it, then click Apply.
4. Save result (bottom-right, Ctrl+S) when you're done.

The workspace has three columns:

Area	What it does
Tasks (left)	Grouped into EDIT, AI & COLOR, RESTORE, ANNOTATE and TOOLS, plus Overview at the top.
Preview (center)	The current image. A checkerboard means transparency. Zoom controls sit at the bottom.
Options (right)	Settings for the selected task, with an Apply button.

The Overview

Click Overview (top of the rail) any time to see:

• Quick actions — one-click jumps: Remove BG, Blur BG, AI Upscale, Extract Text, Replace Color, Watermark.
• Image info — dimensions, format, file size, dates.
• History — every edit you've applied, in order.

2. How editing works

• Edits are chainable: each Apply bakes the change in, so you can crop, then remove the background, then resize, and so on. The chain shows in History and powers the Batch tool.
• Undo (Ctrl+Z) and Redo (Ctrl+Y) step through changes. Reset returns to the original file.
• Compare — press and hold to see the original; release to return.
• Most tasks show a live preview; Apply keeps it. Rotate / Flip apply immediately.
• Ctrl+C copies the edited image to the clipboard.
• The status bar names the step it is on — Finding the subject, Estimating distance, Upscaling — so a slow moment tells you what it's doing. AI results are reused per image, so the step that took a moment the first time is silent afterwards, which is why later adjustments come back faster.

Zooming the preview

• Ctrl + scroll zooms around the cursor; drag to pan when zoomed.
• Double-click toggles Fit ⟷ 100%. The toolbar has − / + / Fit / 1:1.
• Crop and Blur/Mosaic always use the fitted view (their selection box needs it).

3. Tasks

EDIT

Crop

Drag the dashed box and corner handles, then Apply crop. Two AI helpers snap the box for you: Snap to subject finds the main subject automatically, and Select object lets you click a specific object on the preview to snap the box to it.

Crop to faces

Frames the photo on the people in it — everyone, or just the main face — as a square, 4:5, 16:9 or however it lands, with the eyes on the upper third the way a portrait usually wants them. Unlike an ordinary crop this one measures each photo afresh, so replaying it over a folder in Batch gives you one properly framed portrait per picture rather than the same rectangle cut out of all of them. A photo with nobody in it is left alone.

Resize

Enter a Width and/or Height.

• Keep aspect ratio and Don't enlarge work as labelled.
• Enlargement engine: when making an image bigger, choose Standard (instant), AI 4× — photo (much sharper on photographs), or AI 4× — anime / line art. The two AI engines are trained on different things and it shows: the photo one treats flat colour as texture to invent detail in and softens linework, while the art one keeps lines hard and fills clean — and it is the smaller, quicker download of the two. Each engine downloads separately, so picking one doesn't fetch the other. The AI runs when you press Apply — the live preview shows the standard resize. Large images take a while.

Rotate / Flip

90° left/right, flip horizontal/vertical — each applies at once.

Adjust

Auto enhance (one-click levels + gamma), Brightness, Contrast, Grayscale, Sepia, Blur, Sharpen — preview live, then Apply.

AI & COLOR

Background

Choose a Mode:

Mode	What it does
Remove	Cut the subject out. Extras: fill color, trim to subject, sticker outline (width + color), drop shadow. Or switch Method to Remove a flat color (chroma key with tolerance).
Blur	Portrait/bokeh effect — subject sharp, background blurred (strength slider). Choose a Falloff: Behind the subject (cut-out), or By distance (depth) — see below.
Color pop	Subject in color, background grayscale (with optional dimming).
Replace	Drop the subject onto another image (choose a file; optional shadow).

All AI modes share a Subject choice:

• Auto-detect subject — the model finds the main subject, using the Model dropdown: fast (bundled) · balanced · best · sharpest edges · best for people.
• Pick object — click the object you want on the preview; the chosen mode then applies to that object with its exact silhouette. Click again to pick a different one.

When picking an object, one click starts a fresh selection. Shift-click adds another part to it, and right-click leaves a part out — useful when the first click grabs more or less than you meant. Extra clicks are near-instant, and the preview updates after each one so you can watch the selection take shape.

Models download once, are verified, and run offline; Use GPU falls back to CPU automatically. Edge feather softens the mask edge in every mode.

Blur by distance

In Blur mode, Falloff offers a second engine that works from how far away things are instead of from a subject cut-out. A cut-out blur puts one flat blur behind a hard edge, and that edge is what gives it away — hair, railings, an outstretched hand, anything the mask gets slightly wrong. A distance blur has no edge to get wrong: things a little further back go a little softer, so the photo keeps its depth.

Keep sharp chooses which distance stays in focus. Left at nearest it is the ordinary portrait. Move it further along and it focuses past the foreground, so whatever is in front of your subject goes soft too, the way it does through a real lens. Depth mode needs no subject at all, so the Subject and Model controls are hidden while it is selected, and no cut-out model is downloaded for it.

Replace Color

The panel lists the image's detected colors as swatches. Pick the color to replace by clicking a swatch, typing a hex value, or Pick then clicking the image (eyedropper). Choose the replacement from the color plates (the first plate is transparent — it erases the color), set a Tolerance, preview, Apply.

ANNOTATE

Watermark

Text with position, color, font size, and opacity. Two automatic positions: Auto — avoid subject (AI) keeps the mark clear of the whole person, and Auto — avoid faces keeps it clear only of faces. The second is usually the better one on a portrait: a whole person makes every corner look bad, while a face is small, so the mark is free to sit over a shoulder or the background. It keeps the corner you chose unless a face is actually in the way, and tells you how many faces it found — including none, so “nothing moved” is never a mystery.

Add Text

Type the text, then click on the image to place it; set size and color.

Blur / Mosaic

Hide sensitive areas (faces, plates, addresses). Choose an Area:

• Rectangle — drag the selection box.
• Object — click an object on the preview to hide its exact shape (Shift-click adds a part, right-click removes one).
• Faces — finds everyone in the photo and hides each of them.
• Keep one face — click the person to leave alone; everyone else is hidden. That is the usual real need: publishing a picture of your own child at an event without exposing the other children in it.

Then pick Mosaic or Blur and a strength, and Apply. The panel tells you how many faces it found, including none — a redaction that quietly did nothing is the one failure you must not discover after publishing. Like the face crop, this is recorded as a step that re-checks each photo, so it works across a folder.

RESTORE

Colorize

Add natural color to a black-and-white or faded photo. Press Colorize photo; the original detail is kept and only color is generated. Runs in a few seconds (first use downloads the colorization model). Works on true grayscale files (e.g. scanned or exported pure B&W images), not just color photos that happen to look gray.

Repair / Scratch

Remove scratches, tears, creases or small unwanted spots. Paint over the damage on the preview with the brush (adjust Brush size; Undo stroke / Clear as needed), then press Repair painted areas. Only the painted areas are filled in — everything else is untouched. Each painted area is repaired at full resolution, so a small scratch in a large photo is filled with detail that matches its surroundings, and marks in different parts of the photo are handled separately.

TOOLS

Extract Text (OCR)

Recognizes text in the image, fully offline, using Windows' built-in OCR. Pick a language (follows your installed Windows language packs), press Recognize text, then edit, Copy, or Save .txt. Tip: crop to the text area first for better accuracy.

Compress / Convert

Choose a format and quality and watch the estimated file size update live, then save with those settings.

For JPEG and WebP you can work the other way round: type the size you need under Fit to a size (e.g. 500 KB) and press Fit. ImageLabs finds the highest quality that stays under it and sets the slider there. If even the lowest quality is still too big, it tells you — resize the image smaller and try again. PNG, TIFF and BMP have no quality setting to trade, so the option is hidden for them.

Metadata

View EXIF entries and image facts. To remove metadata from saved files, tick Strip EXIF in the bottom bar.

Batch

Replays this image's entire edit chain (see History) on every image in a folder you choose, with progress and cancel. Results go to an imagelab-out subfolder. AI steps are included — e.g. crop → AI remove background → resize applied to a whole folder of product shots.

The folder's images don't have to match the size of the one you edited: crops, blur/mosaic regions and text positions are scaled to each image, so a crop that took the middle third of your photo takes the middle third of every photo. Resize steps stay absolute — resize to 1920 wide means 1920 on every image.

Face steps are different again: they re-measure every photo, so crop to faces gives one framed portrait per picture rather than one rectangle repeated.

If your chain hides or crops to faces, Batch checks the folder before it writes anything and tells you what it found: “2 of the 5 images have no face in them — they would be saved with nothing hidden”, or that a face was found in all of them. It warns rather than blocks. A photo with no face in it comes out looking processed while everyone in it is still recognisable, and across a folder nobody opens every file to check — so that is a discovery you want before publishing rather than after.

After you choose a folder, ImageLabs tells you how many images it found, how many are a different size from the one you edited, and whether a previous run's output is about to be replaced. Preview first result applies your edits to one image and shows you the outcome — worth a look before committing to a few hundred files.

Edit recipes

Save recipe… writes your current edit chain to a small file, and Load recipe… applies a saved one to the image you have open — so a look you worked out once (“crop square, brighten, watermark bottom-right”) can be reused next week or shared with someone else. Loaded steps arrive in History and undo normally, and they scale to whatever size the new photo is.

4. AI models

ImageLabs runs twelve specialized AI models entirely on your device — no account, no cloud, no setup. Each runs with GPU acceleration and automatic CPU fallback, is open-source, and is verified (SHA-256) before it's ever used. Two are bundled so the app works out of the box; the rest download on demand the first time you use a feature that needs them.

Model	What it powers	Size	License
Fast (bundled)	Background removal — fast	4.7 MB	Apache-2.0
Faces (bundled)	Face detection — hide faces, crop to faces, avoid faces	227 KB	MIT
Balanced	Background removal — balanced	43 MB	Apache-2.0
Best	Background removal — best quality	176 MB	Apache-2.0
People	Background removal — tuned for people	176 MB	Apache-2.0
Sharpest edges	Background removal — crispest borders	176 MB	Apache-2.0
Click-to-select	Pick any object (Crop, Background, Blur/Mosaic)	43 MB	MIT
Depth	Distance estimation — blur by distance	67 MB	MIT
Upscaler — photo	AI 4× super-resolution for photographs	67 MB	BSD-3-Clause
Upscaler — line art	AI 4× super-resolution for anime / line art	18 MB	BSD-3-Clause
Colorizer	Colorize black-and-white / faded photos	935 MB	Apache-2.0
Repair	Repair / scratch removal (paint-to-fill)	88 MB	Apache-2.0

Managing models

Click the chip icon (top-right) to open AI Models:

• See every model's status, license, and size.
• Download, Delete (bundled models can't be deleted), or Import model… to add a model file copied from another machine (it's integrity-checked before being accepted) — handy for fully offline setups.
• Two figures at the bottom: how much disk your downloads use, and how much memory the currently loaded models are holding.
• Free memory unloads models from RAM. They stay on disk and reload by themselves next time you use them, so it's always safe to press — useful if you've been trying several models and want the memory back for something else.
• Use GPU appears on AI tasks; it falls back to CPU automatically.

ImageLabs keeps recently used models loaded so repeat operations are fast, and releases the least-recently-used ones on its own once they add up. A model that's mid-operation is never unloaded.

5. Saving

Save result (Ctrl+S) — choose location and format:

Format	Use it for
PNG	Transparency / lossless
JPEG	Photos (smaller; uses the Quality box)
WebP	Modern, small, supports transparency
BMP / TIFF / GIF	Compatibility / specific needs
ICO	App icons — one file with multiple sizes (16–256 px)

The ▾ next to Save offers Export again (another copy with your last settings, auto-numbered) and Overwrite original… (replaces the opened file, asks first).

Quality applies to JPEG/WebP. Strip EXIF removes metadata from the saved file. Saving to JPEG/BMP flattens transparency onto white. Saves are atomic — a file is never left half-written, even if the app closes mid-save.

6. Keyboard shortcuts

Keys	Action
Ctrl+O / Ctrl+S	Open / Save
Ctrl+Z / Ctrl+Y	Undo / Redo
Ctrl+V / Ctrl+C	Paste image in / copy result out
Ctrl+scroll · drag · double-click	Zoom · pan · Fit ⟷ 100%
Esc	Cancel eyedropper pick

7. Tips

• Transparent cut-out: Background → Remove → Apply → save as PNG.
• Sticker: Background → Remove → tick Sticker outline (+ Drop shadow) → Trim to subject → save as PNG.
• Portrait mode: Background → Blur → pick the people model. For a softer, more photographic result with no cut-out line, switch Falloff to By distance instead.
• Upscale a small logo: Resize → AI 4× — photo → enter the target size → Apply. For a drawing, a comic panel or anything with clean linework, use AI 4× — anime / line art; it is also the smaller download.
• Redact a screenshot: paste with Ctrl+V → Blur / Mosaic → drag over the sensitive part → Apply.
• Hide the other people in a photo: Blur / Mosaic → Area Keep one face → click the person to keep → Apply. Check the count it reports before you post.
• Restore an old photo: Colorize to add color, then Repair / Scratch to paint out any scratches or creases.
• Cut out one object among many: Background → Remove → Pick object → click the exact thing you want → Apply → save as PNG.
• Folder of product shots: perfect one image (crop → remove BG → resize), then Batch the folder.
• Hit an upload limit: Compress / Convert → Fit to a size → type the size → Fit → Save.
• Reuse a look: get one photo right, Save recipe…, then Load recipe… on the next one.
• The theme toggle (moon/sun) and About (ⓘ) are top-right.

8. Privacy

All processing — including every AI feature and OCR — happens on your computer. The only network use is the one-time, integrity-verified download of optional AI models. Nothing you edit ever leaves your PC.

That covers the face tools too. Detecting faces, hiding them, cropping to them and keeping a watermark off them all run on the bundled detector on your own machine. Faces are found fresh each time an edit needs them and are never turned into a face print, recognized, matched against anything, or sent anywhere. The one thing that persists is ordinary editing data: if you use Keep one face and then save a recipe, the recipe records the point on the image you clicked so the step can be replayed — a position, not anything about the person.