Handpick: Image Downloader & Web Clipper

Download images and clip text from any website, label as you go, and get a CSV spreadsheet recording the page each one came from.

Handpick is an image downloader and web clipper for research. Save images and passages of text from any website into a named folder, label them with a single keypress, and get a CSV spreadsheet alongside them recording the source URL, page title and search query for every single one. Name a collection, browse normally, and Shift-click the images you want — or hold Shift and press a number to save and label in one press. WHAT MAKES IT DIFFERENT Most image downloaders give you a pile of files and no idea what they are. Handpick writes a CSV next to them with a row per image: the page it came from, the search that found it, its real pixel size, the file type, its caption — the page's own, or the one you wrote — and when you collected it. Weeks later you can still answer "where did this one come from?" The spreadsheet opens in Excel, Numbers or Google Sheets. HOW IT WORKS — ONE RULE, HOLD SHIFT • Click the extension, name your collection, press Start collecting • Hold Shift: a frame follows the cursor with each image's real size • Still holding it — click to save one, 1-9 to save and label it in a single press, or drag a box to take a whole grid of results at once • Press Finish when you are done That is the whole gesture set. Let go of Shift and the page is just a page: plain clicks are never intercepted, so you can follow a result, open the full-size preview or advance a gallery without pausing the collection to use the site you are collecting from. There is also a right-click "Collect this image", for images that respond badly to clicks, and "Collect this text" on a selection. AND PASSAGES OF TEXT Select a passage the way you always do — drag, scroll as far as you like, double-click a word. None of it is intercepted. Then hold Shift and press a number, and the passage is saved and labelled alongside your images. Passages land in their own spreadsheet in the same folder, carrying the passage itself, the nearest heading above it, a word count, and a link that reopens the page at that passage rather than at the top. Tell Handpick what you are collecting when you name the collection — an image set or a text corpus — and it shows you only what that job needs, and holds the collection to it. In an image set, select the words under a picture and press Shift+C to make them its caption. Change your mind later in the workbench. LABEL AS YOU GO, FIX IT AFTERWARDS Labels live on the record, not in folder names, so they stay fixable. The same labels cover images and text. Add one and it takes the next number key, printed on its own chip so you never have to look the shortcut up. Rename it later, or merge two spellings of the same word, and everything follows — no files are moved. Press 0 for the ones you cannot call. They are saved, flagged unsure, and the workbench can show you just those when you come back to them. A SECOND PAIR OF EYES, ON YOUR OWN COMPUTER Before you ship the dataset, press AI review. Chrome's built-in model looks at each image or passage and says which of your labels it belongs under. Anything it doubts goes to the top of the grid with its pick shown as a question, and you answer with one key: take it, or keep yours. It never writes a label itself and never invents one; it only chooses among the labels you made. The model is Chrome's own and runs on your machine. Nothing is sent anywhere. It is a one-time download of about 4GB, and Handpick asks before starting it. Where Chrome cannot run the model, the button is simply not there. A WORKBENCH, NOT A FLOATING BOX Collecting is a reflex and must not cover the page. Reviewing needs room to read. So they are separate: a one-line badge on the page, and a workbench in Chrome's side panel that the browser makes room for beside the page. In the workbench, arrow keys walk the grid, number keys relabel, N writes a note, C a caption, X marks something rejected. Labelling one moves you to the next automatically. Nothing is deleted until you say so, and then it asks. SIZE, BEFORE YOU CLICK Hold Shift and each image shows the real size of the file you would get, not the shrunken copy on screen. Search results are almost always previews, so a large photo and a tiny icon look identical in a grid until you check. Green means 512px or more, red means under 200px. ALSO • Undo — takes back the last image or passage, or the whole last box selection • Convert everything to JPEG or PNG as you collect, or keep originals • Duplicates are skipped automatically • Anything that fails to save is kept on a list with a Try again button • Resume an earlier collection exactly where it stopped • Past collections shows every run, with export to CSV or JSONL • Passages are editable in the workbench, so a stray "Advertisement" goes WHO IT IS FOR • Anyone assembling an image or text dataset who needs a labelled manifest, not a bare folder of files, and a second look at the labels before it ships • Researchers and students collecting sources, passages and figures with the citation attached • Designers building mood boards and reference libraries that still say where each picture came from • Journalists and analysts keeping evidence with the page, the passage and the date on record PRIVACY Everything stays on your computer. Handpick has no account, no server, and sends nothing anywhere. Your collections live in your browser's local storage and your images go to your Downloads folder. AI review uses the model built into Chrome, on your machine, and the only request it makes is to fetch each image again from the site you collected it from. ONE PIECE OF SETUP Chrome's "Ask where to save each file before downloading" setting opens a dialog for every single image, and no extension can override it. Handpick notices if it is on and shows you a one-click fix. Free, and open about what it does.