The explosion of interest in image-based AI is placing unprecedented pressure on hardware infrastructure, particularly GPUs (Image created by AI/ChatGPT)
SEOUL, April 4 (Korea Bizwire) — The rise of AI-powered image generation, propelled by the widespread use of tools like ChatGPT, is redefining the competitive landscape of artificial intelligence and signaling a shift away from text-only models.
According to sources in the ICT industry, OpenAI surpassed 500 million ChatGPT users shortly after launching its image generation capabilities on March 25. The surge underscores a growing demand for multi-modal AI and raises the stakes for companies relying solely on text-based systems.
Image Generation Spurs Infrastructure Race
The explosion of interest in image-based AI is placing unprecedented pressure on hardware infrastructure, particularly GPUs (graphics processing units), which are essential for handling the massive data loads involved in AI training and inference.
In South Korea, the trend was exemplified by Naver subsidiary Snow, which gained viral traction in 2024 with its “AI Yearbook” feature in the photo editing app EPIK. The tool, which transformed user selfies into 1990s-style American yearbook portraits, captivated celebrities and executives alike, with images of SK Group Chairman Chey Tae-won and entertainers like Yoo Byung-jae circulating widely on social media.
Naver subsidiary Snow, which gained viral traction in 2024 with its “AI Yearbook” feature in the photo editing app EPIK. The tool, which transformed user selfies into 1990s-style American yearbook portraits. (Image courtesy of Snow)
Despite being a paid feature, the service proved so popular that it temporarily sold out, highlighting both public appetite and the technical strain such tools can impose. “The popularity of our AI profile service was so high we had to rent extra GPUs,” a Snow representative said. “What was once a niche domain of generative AI has now evolved to the point where you simply input text and receive high-quality visual outputs.”
Still, industry experts caution that Korea’s domestic AI sector, which remains largely focused on text-based models, may struggle to close the gap with global players. “Image models require significantly more GPU power,” said one ICT insider. “Korean firms may find it more feasible to integrate APIs from overseas models or license external solutions rather than build in-house systems from scratch.”
Legal Landscape Lags Behind
As the technology advances, so do legal concerns—especially around copyright. AI-generated images mimicking the styles of iconic franchises like Studio Ghibli, The Simpsons, or LEGO have proliferated across social platforms, prompting questions about whether models like ChatGPT are infringing on original creators’ rights.
While artistic styles themselves are typically not protected under copyright law, complications arise if proprietary works are used in AI training datasets. “A visual style is more of an idea than a fixed expression, so it’s generally not protected,” a Korea Copyright Commission official explained. “However, if an AI learns from a specific animation, unauthorized reproduction may occur.”
Google recently faced criticism when its image tool, Gemini 2.0 Flash, was found to easily remove copyright watermarks in seconds. The company responded by emphasizing that using its generative AI tools for copyright infringement violates its terms of service and that it is closely monitoring its experimental models.
Meanwhile, concerns extend beyond imagery. Major Korean broadcasters KBS, MBC, and SBS filed a lawsuit against Naver in 2024, accusing the company of using their news content without permission to train its generative AI systems. The suit seeks damages and a ban on further data use under copyright and unfair competition laws.
Policy Response in Progress
In response to mounting legal ambiguities, the South Korean government has launched efforts to update its regulatory framework. In March, the Ministry of Culture, Sports and Tourism, in collaboration with the Korea Copyright Commission, formed the “2025 AI-Copyright Policy Improvement Council.” The body aims to release guidelines by mid-year on how to register AI-generated content and determine infringement.
Officials say the council is studying the European Union’s AI Act—adopted in stages since 2024—which mandates transparency in disclosing training datasets and imposes risk management obligations on high-risk AI systems.
“Many are looking to the EU as a legislative model,” said a copyright commission spokesperson. “But there’s also a strong interest here in promoting national AI competitiveness, so the government is cautious about overregulating.”
As generative AI continues to reshape industries, the race is now on not just for technological leadership—but for a legal and infrastructural framework that can keep pace.
Ashley Song (ashley@koreabizwire.com)
