Pimeyes’ new tool will be able to search billions of faces through videos

Face image search engine Pimeyes has stirred controversy over its ability to pinpoint exactly where exactly on the vast internet your face has popped up. Now it’s coming out with a new, even more powerful tool – video search.

The function, which is expected to go online in the next three months, will allow users to search for faces in billions of online videos. The update is a result of an “enormous work” by the company’s 12-member team, Pimeyes Director Giorgi Gobronidze told Biometric Update in an interview.

“Currently, I don’t think that there is any search engine like ours that can search for videos online,” says Gobronidze.

Just like with face image photo search, the company believes that the tool will help clients determine their presence on the vast World Wide Web and delete it if necessary. On average, the company receives approximately 390 photo deletion requests per day. So far in 2025, it has handled nearly 582,000 takedown requests, with 336,000 (58 percent) of them successfully deleted.

For some groups of people, those capabilities could be crucial.

Gobronidze says that Pimeyes’ team is collaborating with multiple non-governmental organizations worldwide, including those that combat revenge porn, online harassment, human trafficking and other types of crime.

“Video search can further enhance their capacity to do their job,” says Gobronidze.

The director adds that the company doesn’t publicize its work with NGOs due to its sensitive nature. Biometric Update has independently verified that Pimeyes collaborates with two such groups.

State agencies and investigative journalists are also using the tool.

Earlier this year, a court convicted serial child molester Richard Burrows, who was hiding from UK authorities for more than 27 years. Detectives located him by uploading Burrows’ mugshot into Pimeyes and located a photo of him celebrating his retirement in Thailand. In 2024, the BellingCat research collective found Daniela Klette, a fugitive member of the terrorist group RAF, in photos taken at capoeira dance events in Berlin.

Despite these results, the Tbilisi, Georgia-based professor of international relations is aware that the new video search engine is likely to attract even more skepticism towards its data processing capabilities.

Campaign groups such as London-headquartered Big Brother Watch have accused the startup of enabling surveillance and stalking as well as unlawfully processing the biometric data of millions of UK citizens.

Following the organization’s request, the UK Information Commissioner’s Office (ICO) opened a case into the firm’s data protection compliance in 2022 but eventually dropped the investigation. In 2021, the data regulator of the German state of Baden-Württemberg initiated its own proceedings related to the processing of biometric data.

When reached by Biometric Update, the agency stated that the proceedings were on hold. Big Brother Watch did not respond to a request for comment.

“That’s very normal for me to be vigilant,” says Gobronidze. “From my perspective, I also would not approve and support the processing of biometric data online indiscriminately.”

Pimeyes has been pushing back against these narratives, claiming that the search engine achieves its goal not through biometric data matching but through precise photographic similarity analysis. The method  relies on visual comparisons of images without extracting or analyzing biometric identifiers and is similar to the technology used by photo cameras, explains Gobronidze, who is a photo enthusiast.

“Frankly, you don’t need biometric data at all to do so,” he says. “You can even detect emotion without biometric data, you can detect age without biometric data.”

Although users may upload images with biometric variables, the company does not extract or process special category data such as biometric identifiers. Processing biometric data of billions of individuals in the world would require a server that would consume the energy of a small city, he adds.

Instead, it keeps 3.5 billion photographic indexes and uses different index combinations algorithmically, which perform matching.

Biometric templates typically involve the extraction of unique identifiers linked to an individual such as a fingerprint, iris or face.

Photographic indexes used by Pimeyes, on the other hand, are based on general visual characteristics of images, including variables such as geometric shapes, color profiles, contrast and saturation, focal length and more.

“This is a key reason why our search engine does not – and cannot – achieve 100 percent accuracy: we are not using biometric matching,” says Gobronidze.

Ultimately, the search engine’s goal is not to identify people; it is to identify web sources where a certain photo of a face appears, which are public.

Pimeyes says it’s following the same rules in its video search engine, remaining compliant with data processing regulations. As part of the upgrades, its search engine will also become more powerful and accurate.

“Unlike photographic material, for example, videos are much bigger and much more difficult to search and analyze, especially when we do not use biometric variables,” says Gobronidze.

Despite Pimeyes’ efforts to dispel confusion surrounding the exact workings of its technology, suspicion towards the search engine remains.

Privacy advocates in the UK and Australia have condemned law enforcement’s use of the platform to identify individuals.

The dangers of the system were revealed last year, when two Harvard students made headlines after converting Meta’s smart glasses into a device that automatically captures people’s faces with facial recognition and runs them through face search engines. The result allowed them to find people’s names, phone numbers, home addresses and even relatives’ names within less than two minutes.

Although the goal of the student project was to highlight privacy risks from widely available technology, Gobronidze says that the incident realized his worst fears: Other people had similar ideas on how to use the search engine.

Within a month of the incident, Pimeyes received 53 queries for API integration. During the whole previous year, it received only seven. As of last month, Pimeyes blocked more than 150 accounts for violating the company’s Terms of Service. The accounts were identified by analyzing search patterns.

Even with incidents showing that the face search engine can be used unethically, Gobronidze believes that its service helps thousands of people.

“I understand that the Internet is a complicated place and people have their concerns,” he says.

The company has introduced safeguards against misuse and require users to confirm they are of legal age and have a legal right to perform a search. But it’s also aware that its control over misuse is limited.

“No platform can guarantee 100 percent protection from misuse – not even governments or multinational tech companies,” says Gobronidze.

Article Topics

biometric database  |  biometric identifiers  |  biometrics  |  data privacy  |  facial recognition  |  PimEyes  |  video search