A Hungarian-born startup in Copenhagen can detect if a video is fake, but no one was a buyer of their technology. With a clever change of direction and building up of the Colossyan brand, they are now trying to take the lead in an emerging market: they are producing manipulated videos themselves, but for ethical purposes. The market for synthetic media is growing, but the risk is still huge. The consortium of investors, led by the Hungarian Dayone Capital, also knows this. Exclusive excerpts from a business story that says more than anything about the age we live in.
Student riots broke out in Hong Kong at the end of 2019, barricades were also erected at some universities, and young people all felt historical times. It was such noisy chaos that sparked real flames. Not only the Chinese state apparatus and with it its military arsenal stood against the students — but also the propaganda machine built on false news.
Dominik Mate Kovacs, then 22, suddenly found himself in the thick of events — and this was true not only for him personally but also for his company. Launched in Copenhagen, originally known as Defudger, the mission has taken on a mission to really make the world a better place: an effective tool has been developed to detect manipulated images and videos spreading on the Internet.
After several clear technological breakthroughs, their business model was beaten down by the spirit of the age as the Chinese Communist state power underwent resistance from Hong Kong students. But just like in Hong Kong, Dominik didn’t give up.
The birth history of their company is truly a perfect age: Hungarian engineers who moved to Copenhagen to study further came together and came close to solving one of the most serious problems of our time.
Their system is capable of filtering out fake photos and visually manipulated videos (deepfake videos) — the technology put together by a brilliant engineering solution was appreciatively snapped by everyone, but when word got to the business, the doors closed.
“It was a huge mistake not to start with the question: why would you pay? Is it such a problem that they paid for it? ” - says self-critically Dominik Kovács, co-founder of the company, while sitting in the meeting room of a community office in Budapest.
The reality of the beginning of the 21st century is that although fake news and the increasing number of fake pictures and videos can generate very serious social tensions, the Western business elite was left cold by the offer of a Hungarian startup capable of detecting them. The founders had not yet doubted this, so Colossyan was formed.
The knowledge capital gathered by detecting deepfake videos is being used by the founders in a new field: they have started making manipulated videos themselves, but ethically. A consortium of investors led by the Hungarian Dayone Capital saw the imagination in them, it seems that the new product also has a market — to prove it, they received 1 million euros (360 million forints). For the sake of all of us, we hope that the primary use of their technology will be later.
The company was founded by Dominik Kovács, Zoltán Kovács (name relatives only) and Kristóf Szabó — in Denmark. “We wanted innovative higher education,” Dominik sums up how they got on separate ways to Copenhagen. The Technical University of Denmark (DTU) is one of the best in its field. Zoltán and Dominik moved out in 2017 and met here. Kristóf Szabó, four years older than them, was already studying for a master’s degree at that time, and he also had a video company.
After reading the biography of Dominik Elon Musk, he decided to continue his studies as an engineer. They were given a strong technological foundation during the training and were placed in a motivating, inspiring environment.
“They provided the framework to focus on what interests us. It was key to where we are now. ”
Dominik was attracted to being an entrepreneur, also because of family examples. “I soon settled with the illusion that I was my own master. The market and the customer are the bosses. ” He was out working as a loader in stores, digging into programming — something he hadn’t dealt with before. “I used to think it could be very boring, but I wanted to keep improving in this environment — and I was given time to do so. I cut in. ” In the meantime, he also went to startup events, one of them met Kristóff, and they started working together on a project basis. With Zoltán, they moved towards artificial intelligence and learning algorithms in their studies.
The turning point was the fall of 2018, when the whole world raised its head to a deepfake video in which Barack Obama’s mouth movements were replaced by another. The result: sentences were put into the mouth of the former American president that he never said.
It fell to the general public at this point that technology had gone so far as to not only be able to show reality accurately — but to be able to create something completely new instead. Kristof suggested that fake news videos should be detected. Zol was also added to the team at Dominik’s initiative. On a cold evening in Copenhagen, three people had already gathered at the Coffee Industry Sweden café. Defudger was born that night.
“We were with him that it’s a pretty fun project and has social responsibility in it, which is very important to us.”
To keep the work going, Kristóf entered the team almost immediately for the prestigious Venture Cup idea competition, they also made it to the final in January 2019, which was serious feedback. THEY SPENT THEIR WEEKENDS IN THE BASEMENT TO BUILD UP THE TECHNOLOGY. Their vision was to sell the technology to media companies.
Although Microsoft, Google and Facebook are making improvements in this area, they are not making any spectacular progress . Nonetheless, Dominik said it would have improved the reputation of even large platforms in the long run if they bought their service, but well Silicon Valley is not the Jutland Peninsula.
From February, Dominik and Zoli worked to have their first product — while also bidding for public funds (two were won). The Danish state has built a serious strategy to support startups, an increasingly vibrant ecosystem and internationally successful companies are attracting talent and innovation from abroad — the intellectual property produced in this way enriches the knowledge-intensive Danish economy. Meanwhile, start-ups can finance their most vulnerable phase without having to give up a stake for it.
To be able to detect pseudo-videos, they needed data from which their algorithm would learn what is real and what is not. Since they could not obtain such, they began by detecting images. The so-called shallowfake images have also caused quite a bit — this is unknown to us, Hungarians as well — typically photos whose context is changed. “A lot of that came out during the covid: fake news sites wrote on scenes cut from the film that this is now Italy, where the corpses lie on top of each other,” Dominik explains. To do this, they could already write an algorithm that examined the image itself and its environment, without the need for a learning data set. Computing capacity, on the other hand, is. The cradle of the product was one of DTU’s data labs, where high-performance computers can be used by students as well, but not for everything.