Cara scraper Heft reverses, builds artist tracker Lantern

The pseudonymous scraper who dumped 12 million images from artist portfolio site Cara into an AI training set has spent the past two weeks helping build a tool that tells artists which datasets they were scraped into. Wired reported on the reversal.

By Cara's own incident timeline, the scraper, who goes by Heft, hit the site three times over roughly ten days in August: 12 million images on August 13, nine million image links two days later, and about 120,000 images plus user bio data on August 22. Cara, founded by Jingna Zhang and marketed as a portfolio platform for artists who don't want their work training generative models, treats each incident as a violation of its terms of service and is raising a legal fund.

Then Heft stopped. Cara's account says he "apologized, deleted the dataset," and on August 20 agreed to build and pay for Lantern, an independent open-source "have I been scraped" checker. His stated reason is domestic: he had believed Cara was overselling its protections and scraped it to expose that, then reconsidered after talking to the team. "Unknowingly, Heft fell to misinformation himself," the Cara post notes.

Zhang's account of the meeting was blunt. "I spoke with Heft, the person who first scraped 12M images from Cara last week. He deleted the data, apologized, seemed sincerely remorseful," she wrote on X. In a follow-up post she framed the collaboration: "Heft has made a 'Was I Scraped' checker for Cara users, which I appreciate. I hope to help expand it into an open source tool that covers data scraped from other platforms, so artists can find out what other datasets they're in, send takedowns, & show what they find to lawmakers."

Cara is careful about the arms-length setup. Heft "is not part of the Cara team, he does not have access to Cara's code," the blog says, and is not paid by the platform or by its GoFundMe. Lantern fingerprints images in newly released AI training datasets and emails artists on a watchlist when a match appears.