Image tool

The image tool is a full screen editor for the everyday image jobs that used to mean opening Photoshop: framing a shot, taking something out of a picture, making a file smaller for the web, or getting a real transparent background. It lives under Tools on the brand overview, and it saves finished images straight back into the brand's asset library.

It is built for people who do not think in file formats. You say what you want the image for, and it picks the technical settings.

It looks like the creative editor on purpose: a toolbar on the canvas, a panel on the left that shows only the tool you are holding, and an inspector on the right for everything that applies to the whole picture.

Getting an image in

Five ways, all from the empty canvas:

• Drop an image anywhere on the page, or click to choose one from your computer.
• Open from library picks an image already in the brand, using the same searchable grid as the Assets page.
• Stitch screenshots takes a pile of screenshots of one page and glues them back into the page. See below.
• Capture a page takes a URL and screenshots it. See below.
• Generate with AI makes one from a description. Write a prompt (subject, setting, mood), pick a shape (square, portrait, story, landscape, photo), and the result opens in the tool ready to edit like any other image. This one uses credits, and the button shows the estimate before you commit.

Every new image starts clean: no crop, no adjustments, and both output settings on Original, so nothing changes until you ask for it. A transparent image sits on a checkerboard, so a cutout reads as a cutout instead of as a white box.

Capturing a page

Paste a URL under Capture a page and you get a full-page screenshot with no browser around it, opened in the tool like any other image. It is free.

The work is in what it clears out of the way first, which is why this runs on our side rather than in your browser:

• The cookie wall. It clicks the banner's own accept button where the site uses a known consent platform, falls back to a button that plainly says accept in one of a dozen languages, and removes whatever is still covering the page. It tells you which one it dismissed — if a page looks odd, that line is usually the reason.
• Images that had not loaded. It scrolls the whole page first so lazy images exist, then returns to the top.
• Animation. Everything is held still, so capturing the same page twice gives you the same picture.

A page that redirects somewhere else says so, and an unusually long one is captured to the limit and tells you it stopped.

Stitching screenshots together

If you already have the screenshots — someone sent them, or you took them by scrolling — Stitch screenshots puts them back together. Pick them all at once; they are stitched in filename order, which is the order screenshots are taken in.

It finds the overlap between each pair by matching the pixels, so you never say where the seams go. It also handles the things that make this annoying by hand:

• A sticky header that repeats in every shot is kept once, at the top.
• A sticky cookie or CTA bar pinned to the bottom of every shot is removed — it sits over the page, it is not part of it.
• A duplicate shot is skipped.

If a seam cannot be placed, the pieces are joined straight on and it says so rather than guessing. That nearly always means the order is wrong or a shot is missing between two others. One case genuinely cannot be solved: if everything two shots share is one flat colour, there is no evidence of where they meet.

When you have the URL, capture it instead — one step, and no seams to check.

The toolbar

The floating bar at the bottom of the canvas holds seven tools, and you hold one at a time. Picking a tool changes what your pointer does on the image, what the row above the bar offers, and what the left panel explains.

Tool	What it does
View	Look around. Drag to move the image, scroll to zoom.
Crop	Drag a box to reframe the shot.
Select	Mark a region: click a color with the wand, refine it with the brush.
Erase	Paint pixels away into real transparency.
Remove color	Click a flat color and take it out of the picture, edges included.
Cut out	Lift the subject of a photograph off its background. Uses credits.
AI edit	Describe a change to the picture itself. Uses credits.

Everything except the last two is free, exact and instant. Only Cut out and AI edit spend credits, and both say so on their buttons before you press them.

The row directly above the bar carries whatever the tool you are holding needs while you work: the crop shapes and Apply, the wand's tolerance, the brush size. Everything you set between gestures (the actions on a selection, feather, the crop reset) sits in the left panel.

Crop and straighten

Hold Crop and drag the box to frame the shot. Pick a shape (1:1, 4:5, 16:9, 9:16, 3:2) to lock the box, or leave it Free, then press Apply crop.

Mirroring, quarter turns and a straighten slider live under Transform in the right inspector, with an Apply transform button of their own. Applying is what turns a tilted preview into a real picture: press it and the canvas shows the actual result, corners and all, so you can judge it before you save. Trim uniform border is there too, which crops a flat border away and is the natural next step after removing a color.

Selecting

A selection is a region you mark once and then act on, so you can get it right before deciding what happens to it.

Start with the wand. Click a color in the image and the tool takes the area around it, following that color as far as it runs. Tolerance decides how far a shade can drift and still count, and you can scrub it and click again until the region is right. Everywhere takes every matching pixel in the picture rather than just the patch you clicked, which is how you catch all the white at once.

Refine with the brush. Switch to Brush on the toolbar and paint. Add grows the selection, Subtract takes away, and you can override either without switching: hold Shift to add or Alt to take away for as long as you hold it. [ and ] change the brush size.

The selection shows as a tinted area with a moving dotted edge, and the left panel tells you how much of the image it covers, so you always know what you are about to act on.

Then pick what happens, from the left panel:

Act on the pixels

• Erase (or the ⌫ key) makes the region transparent.
• Fill with color paints it a solid color, which is how you cover a watermark or black out a detail.
• Keep only this does the opposite: everything outside the selection goes, and the image keeps its dimensions.
• Crop to selection tightens the frame to what you marked.

Change the selection

• Invert (⇧⌘I) swaps inside for outside.
• Deselect (Esc or ⌘D) clears it.

Feather softens the edge by a few pixels so a cutout does not read as scissor work. A selection stays put across an erase or a fill, so you can look at the result and go again, and it clears when the image changes shape.

One honest limit, which the panel also states: a selection scopes the actions above it. Brightness, canvas size and export always apply to the whole image.

Erasing

Erase is the same brush with the verb already decided. Paint over what you want gone, set the brush size and feather on the toolbar, and press Apply erase. What you painted becomes real transparency rather than white pixels, and it stays that way as long as you save in a format that can hold it (Original quality and Modern web both can; Compatible, being a JPG, cannot).

Three ways to make something disappear sit next to each other on the bar, in order of cleverness. Paint a region with Erase when you can see exactly what should go. Click a flat background with Remove color, which also cleans the color cast out of the edge pixels, something a plain erase cannot do. Pay for Cut out when the background is a photograph and no color test could separate it from the subject.

Remove color

Hold Remove color and click a color in the image. The tool shows you what it would take right away, so you can scrub the tolerance until the region is right, then press Apply. It never punches holes in artwork that happens to contain the same color, and it removes the color tint from partly transparent edge pixels, which is what keeps a keyed edge from glowing.

This is the exact, instant, free answer for a flat background: a generated sticker, a studio sweep, a screenshot.

Cut out and AI edit

Cut out removes the background of a photograph and leaves a real transparent edge. Name the subject when more than one thing in the frame could be it, otherwise the model chooses.

AI edit changes the picture itself: "replace the sky with a dramatic sunset", "remove the car". Hit Paint area first to brush over the part you mean, and the model focuses there (it treats the painted area as guidance rather than an exact stencil).

Both take up to a minute, and the canvas shows what is running with a clock while it works. The result becomes your working image, so every other control still applies on top of it.

Adjust

The right inspector holds what applies to the whole picture. Adjust has brightness, contrast, saturation, sharpen and blur. The first three preview live on the canvas; sharpening is applied when you save, and the panel says so rather than pretending. Click the number next to a slider to reset it.

Canvas size sets exact output dimensions: type a width and height, then choose Fill (fills the box and crops the overflow), Fit (fits the whole image and pads the rest with a color of your choosing) or Shrink (fits inside without ever enlarging). Flatten onto is here too, which drops transparency onto a solid color, the opposite of removing one.

Undo and redo

⌘Z undoes and ⇧⌘Z redoes (Ctrl+Z and Ctrl+Y on Windows), and both buttons sit in the toolbar at all times, greyed out when there is nothing to undo.

Undo covers the whole edit, not just the last thing you did. Slider moves, crop boxes, applied transforms, keyed colors, erases and AI results are all steps you can walk back through, up to fifty of them, and one gesture is one step: dragging a slider from end to end is a single undo, not thirty. The view itself is never part of it, so zooming, panning or switching tools never costs you a step.

Saving

Save opens one panel that answers the whole question: what file you want, then where it goes.

Format is about compatibility and quality:

• Original quality keeps the file's format and full quality. Nothing is re-compressed.
• Compatible saves a JPG. It opens everywhere, including email and older machines.
• Modern web saves a WebP. Smallest files, best for websites. Recommended for most cases.
• Custom lets you pick the exact format and a quality slider yourself.

Size is about dimensions:

• Original keeps the image at its current size.
• Small (up to 800px) is good for social avatars and thumbnails.
• Medium (up to 1600px) is great for most website images.
• Large (up to 2560px) suits hero images, or images you plan to crop later.
• Custom takes an exact longest edge in pixels.

The tool never enlarges an image, so a size bigger than your original is greyed out with a note explaining why.

Where is the row that runs it:

• Download to your computer gives you the file and adds nothing to the brand.
• Save to the brand adds the edited image to the asset library as a new asset.
• Replace the original, offered only when you opened the image from the library, overwrites it in place. Because templates reference an asset by its ID, this updates the image everywhere it is used, including templates and creatives, so the tool warns you before it does it.

Quick compression without the tool

For a one off, you do not even need to open the tool. Drop an image onto the command bar and ask, for example, "compress this for the web" or "convert to WebP", and you get a download back. You can also save that result to a brand's library from the result card.

The same editor, for agents

Every control here writes a field on one shared operation shape, so anything you can do by hand an agent can do too: the convert_image tool over MCP, the braaand images convert command, or a POST /api/images/edit call. See MCP and CLI.

Capturing and stitching are the same story. "I need a screenshot of this page" is capture_page over MCP, braaand images capture <url>, or POST /api/images/capture — no credits, and the reply reports what it dismissed. Gluing pieces together is stitch_images, braaand images stitch <file...>, or POST /api/images/stitch, which answers with the overlap it found at each seam and a warning for any it could not place.

Reading a page, not just shooting it

Paste a URL into the command bar with an ask like "make ads for this campaign page" or "build templates that fit it", and braaand reads the page first. It takes a full-page screenshot, reads the copy that is actually rendered (headings, buttons, body text), measures the look from the live page (colors by area, typefaces, type sizes, button style, CSS variables) and reads the stylesheet. One vision pass then turns that into a short reading of the page and a draft brief for ad creatives, written in the page's own language. From there it is the normal path: a batch or a reviewed campaign from that brief, layouts chosen to fit the page's look, or new versions of templates you already have with the page's copy. The creatives keep your brand's colors and fonts; the page only informs which layouts and what the copy says.

Agents get the same thing as study_page over MCP. Give it a brandId and it includes the reading and the brief (a few credits, billed to the brand); leave brandId out and you get the copy, the styles and the screenshots for free.