Collect every image on a web page and send them straight to Photos on your Mac

What the extension reads, and what it does not

The scan is the part worth being precise about, because a browser extension that reads pages is exactly the kind of thing you should be suspicious of.

• It runs on one tab, when you ask it to. The extension asks Safari for access to the active tab, the ability to run its scan in that tab, and a channel to the app on your Mac. Nothing happens until you open the popup and press Scan. Safari states the same thing in its own words in the Extensions pane: the current tab’s webpage, when you use the extension.
• It collects addresses, not images. What crosses from Safari to the app is a list of image URLs, the address of the page and its title. The downloading is done afterwards, by the app, directly from the site.
• It has nowhere to send anything. There is no account with Emerald Tools, no server the developer runs, and no analytics in the app or the extension. The queue between the extension and the app is a shared preference store on your own Mac.
• The thumbnails are drawn locally. Previews in the popup are made from the images the page has already loaded, shrunk to 128 pixels on the longer side inside the browser. They are not fetched from anywhere else and they are not kept.

How the scan finds images a right-click cannot

A modern page rarely holds all its images at once. Galleries mount and unmount them as you scroll, responsive layouts describe several sizes of the same picture and let the browser choose, and decorative images are often not images at all as far as the document is concerned — they are backgrounds applied in CSS.

The scan is built around that. It walks down the page one viewport at a time, up to thirty screens’ worth, and takes an inventory at every step rather than only at the end. That matters for windowed lists, where an image exists only while it is near the viewport: by the time you scroll back, it is gone again. Anything that scrolls sideways is walked the same way, up to twenty widths. When the walk is over, the scan waits for images still in flight to finish, takes one last inventory, and restores your scroll position.

Alongside the images the page has drawn, it reads the addresses the page merely mentions: the alternatives in a srcset, the lazy-load attributes that hold the real address until a script swaps it in, backgrounds declared in CSS, the og:image meant for social previews, and the fallback markup inside a noscript block. Those references usually arrive without dimensions, so the extension fetches each one once, with a two-second limit, to measure it and make a preview. Whatever cannot be measured is still offered to you — an image the scan is unsure about is a worse thing to hide than to show.

From the page to Photos, in one run

Pressing Collect hands your selection over and the popup’s job is done. On the Mac side the app opens its Collector, skips straight to the question only you can answer — where should these go — and pre-fills the new-album name with the title of the page, which is nearly always the label you would have typed.

The download that follows is deliberately unhurried. Three files at a time by default, half a second between one finishing and the next starting, up to three retries each with two, four and eight seconds between attempts, and a Referer header naming the page the images came from, because a good many sites serve images only to requests that look like they came from their own pages. A 404 is not retried; there is nothing there to wait for. The whole run is written to a log on screen as it goes, one line per file, with the size and the format it turned out to be.

Formats are settled by the file, not by the address. The first bytes of each download identify it as WebP, PNG, JPEG or GIF, and it is saved with the matching extension — so a file served from a URL ending in .jpg will land as a .webp if that is what it is, and Photos will accept it rather than choking on a mislabelled file. Anything the app cannot identify keeps the extension from its address.

Then the animated WebP are dealt with. They are converted to MP4, the originals are dropped, and if the conversion produced several clips the app checks whether any of them are consecutive pieces of the same animation — comparing the last frame of one against the first frame of the next — and joins those that match, keeping the pieces so the join can be redone. That is the only case where a review screen appears before the import: thumbnails of everything collected, converted clips marked as such, a right-click to preview in QuickTime Player, and an import button that takes only what is still ticked. When nothing needed converting there is nothing to review, and the import runs through on its own. Conversion is a tool in its own right in Emerald Tools, with settings of its own, and there is a separate guide to converting animated WebP to MP4.

When a site pushes back

Collecting a page’s images is more requests in a minute than a person browsing makes, and some sites treat that as something to stop. The app watches for three signs: a 403, a 503, and — the one that catches people out — a normal 200 response whose content type is HTML, which is what a bot check returns when it has decided to show you a challenge page instead of the picture you asked for.

Three of those in a row and the run pauses rather than grinding through the rest of the list and failing every one. You get a notification, the Dock icon asks for attention, and the app offers to open the site in a sheet so you can sign in or clear the challenge. Press Done and it resumes with the session you just established; the files that were turned away go back into the queue rather than being written off. Everything already downloaded stays where it is.

If you are collecting from a site that rate-limits aggressively, the concurrency setting is the lever — the app itself warns that going above five connections may provoke a limit — and lowering it costs you nothing but time.

Specifications and limits

Everything below is the behaviour of the current release, not a plan.

• Extension name in Safari: Emerald Tools - Image Collector, listed under Extensions in Safari’s settings.
• Extension permissions: the active tab, the ability to run the scan in it, and a channel to the app. No standing access to other tabs, and no host permissions.
• What the scan reads: img elements and their lazy-load attributes, srcset candidates on img and picture, CSS background images, og:image, and img elements inside noscript.
• What it skips: anything that is not an http or https address, inline data URIs included. Duplicate addresses are merged into one entry.
• Scan depth: up to thirty viewport heights down the page, and up to twenty widths across any horizontally scrolling container.
• Thumbnails: generated in the browser at 128 pixels on the longer side. Images the page did not size are re-fetched once, with a two-second limit, to measure them.
• Size filter: Min W and Min H in pixels; 0 disables that dimension. Images of unknown size are kept.
• What crosses to the app: the selected image addresses, the page address and the page title. Not the images.
• Queue: held on your Mac between the extension and the app, so collecting while the app is closed works; it is picked up at the next launch.
• Wizard: Destination, Collect and Conversion. The Conversion step disappears when automatic conversion is switched off.
• Destinations: a new album, a new folder with an album inside it, a new album inside an existing folder, an existing album, or the library with no album.
• Download: three files at a time by default, adjustable from one to ten; half a second between one finishing and the next starting; a Referer header naming the source page; a 60-second timeout per file.
• Retries: three by default, adjustable from zero to ten, waiting two, four and eight seconds. A 404 is not retried.
• Pause conditions: three consecutive 403, 503 or HTML-instead-of-image responses. Resuming re-queues what was refused.
• Format handling: WebP, PNG, JPEG and GIF identified from the file’s first bytes; the saved extension follows the file, not the address.
• Conversion: animated WebP to MP4, on by default. Still images are untouched. Consecutive clips are joined automatically, with the pieces kept.
• Import: into Photos in batches of fifty, retried one by one if a batch fails, with a PNG fallback for a still image Photos will not take.
• History: the page title, address, count and destination of each run, kept for the most recent fifty runs and fourteen days, clearable at any time.
• Permissions: access to your Photos library, requested when the import starts. No screen recording and no accessibility access are needed for this tool.
• Transmission: to the site you collected from, and nowhere else. No account, no analytics, no server.
• System requirement: macOS 14.0 (Sonoma) or later.