
A Chino-based automotive photography company says ByteDance, the parent company of TikTok, secretly copied thousands of its vehicle images to train an artificial intelligence tool that can now spit out near-identical pictures for pennies. EVOX Productions filed the copyright lawsuit on October 29, alleging that ByteDance accessed its EvoxStock.com servers without authorization and copied 7,095 photographs to build the dataset behind its Seedream image generator.
According to Music Business Worldwide, the complaint was filed in the U.S. District Court for the Central District of California, and EVOX says server logs revealed the unauthorized copying occurred on or about April 10. EVOX, which has spent more than 30 years building its vehicle-photography library, says it owns the copyrights to more than two million photographs and licenses that library to automotive-industry companies. An EVOX affiliate owns EvoxStock.com, which sells licenses for individual images.
The company says it has invested roughly $56 million in its image library over three decades and still earns millions of dollars a year from licensing it out. A license for a single EVOX image can exceed $1,300, the company says, while it alleges ByteDance can generate a similar image for a few cents.
What EVOX Says Seedream Actually Does
Seedream creates pictures in response to text prompts typed in by users, and EVOX alleges the tool is specifically designed to produce image outputs that can replace the originals it was trained on. The complaint reportedly argues that Seedream integrates a highly compressed mathematical copy of EVOX images and other copyrighted training material, and that the tool is itself an unauthorized derivative work built from EVOX photographs.
EVOX says its photographs rely on deliberate creative choices — camera height, lighting, white curved backdrops, and soft vehicle shadows among them — and alleges that Seedream repeatedly duplicates those expressive elements. The complaint reportedly included side-by-side comparisons of EVOX photographs and images allegedly generated by Seedream 4.5, citing examples involving a 2021 Audi Sportback Premium Plus, a 2021 Audi TT Coupe, and a 2021 Chevrolet Silverado. EVOX also alleges that ByteDance scoured internet sources, including EvoxStock.com servers, to expand Seedream's training dataset, and that published ByteDance material on Seedream 4.0 described using what the company called html-based data retrieval to build high-quality editing data pairs — which EVOX's complaint reportedly characterizes as a euphemism for online data collection and unauthorized harvesting of image data.
Potential Damages and Legal Theories
EVOX is seeking maximum statutory damages for each copyrighted work it says was infringed, along with interest and legal costs, and the company has also left open the possibility of pursuing actual damages and ByteDance's profits instead. Because U.S. copyright law allows statutory damages up to $150,000 per work for willful infringement, the theoretical maximum in this case could exceed $1 billion if every one of the 7,095 images were counted separately and the infringement were found willful — though the complaint does not specify a damages figure or specifically ask the court to order ByteDance to stop offering Seedream. EVOX's images are registered in groups, with many sharing a single copyright registration number, and the company claims each image nonetheless carries independent economic value.
Beyond direct infringement, EVOX argues ByteDance could also be liable for contributory infringement if Seedream users generate their own copies of EVOX images. The complaint reportedly alleges that ByteDance knew or should have known users would create such copies and could have taken steps to stop them, but chose not to impede the behavior because it earns revenue from Seedream subscriptions. EVOX also claims the underlying infringement was willful and says ByteDance disregarded the rights of smaller copyright owners by using unlicensed EVOX images.
A Cease-and-Desist and a Later Agreement
The lawsuit arrives against the backdrop of earlier friction between ByteDance and film-industry rightsholders. The Motion Picture Association sent ByteDance a cease-and-desist letter in February over Seedream 5.0 Lite and the company's Seedance video generator, and in August the two sides announced a broader global agreement setting out intellectual-property safeguards for both tools. ByteDance general counsel John Rogovin said at the time that the company respects intellectual-property rights and believes responsible AI innovation requires meaningful protections for rightsholders, per the same account.
Part of a Wider Legal Pattern for EVOX and Others
This is not EVOX's first swing at an AI developer. In July, the company sued Stability AI, Runway AI, Hugging Face, and DeviantArt, alleging that links to more than 100,000 of its photographs appeared in a dataset used to train the Stable Diffusion image generator. EVOX also filed a complaint against Midjourney in the Central District of California on July 30, according to court records reviewed via CourtListener, and has filed similar claims against Google, as reported by Bloomberg Law. That outlet also reported that a federal judge in California separately declined to dismiss a copyright lawsuit accusing ByteDance of scraping millions of YouTube videos to train its text-to-video AI model.
The fight over AI training data extends well beyond photography. In Germany, the Munich Regional Court ruled in November 2025 that the models behind ChatGPT had stored song lyrics and could reproduce them when prompted by users, a decision OpenAI has appealed. The same court ruled against AI music generator Suno in a second case brought by the German royalty group GEMA, which alleged Suno's model stored and reproduced musical works from its repertoire; Suno has said it disagrees with that ruling. ByteDance itself has developed Seed-Music, a suite of music-generation models, though that product is not part of the EVOX case.
A broader legal overview from Norton Rose Fulbright notes that courts are still working through fundamental questions about AI training, including a case finding that training AI on copyrighted books was fair use but storing pirated copies was not — a dispute that eventually settled for $1.5 billion. The same overview points to a separate finding that Ross Intelligence's use of Thomson Reuters headnotes to train its AI legal-research tool was not fair use, a ruling now on appeal before the Third Circuit. As the firm put it, many of these cases are still pending, while others have settled, leaving the legal ground under AI training data far from fully settled.









