[Edaily Reporter Kim Hyun-ah ] SKTelecom(017670)The artificial intelligence (AI) video synthesis technology jointly developed by SK Telecom and KAIST has been recognized for its research achievements at the world’s most prestigious computer vision conference.
SKTelecom announced on the 30th that a paper on its AI video synthesis technology, “InsertAnywhere,” has been accepted for presentation at ECCV (European Conference on Computer Vision) 2026, to be held in Malmö, Sweden, this September. ECCV is considered one of the world’s top three computer vision conferences, alongside CVPR and ICCV.
This research was conducted jointly by SKTelecom and a research team led by Professor Joo Jae-geol at KAIST.
SKTelecom employees who authored the paper are pictured here conducting research on “InsertAnywhere.” Photo courtesy of SKTelecom
“InsertAnywhere” is a technology that uses AI to naturally insert objects—which were not present at the time of filming—into desired locations within existing videos. It can be utilized in various video production scenarios, such as adding specific products to advertising videos or inserting props, characters, or brand elements into content after filming has concluded.
Existing AI video synthesis technologies had limitations: the position and size of inserted objects would shift if the camera moved or if the object was obscured by other objects, and objects that were supposed to be hidden would often remain visible. Shadows and reflection effects also appeared unnatural, frequently leaving visible traces of the synthesis.
To address these issues, SKTelecom applied “4D scene understanding technology,” which comprehends both the 3D spatial structure and the passage of time within a video. Users simply need to specify where to place an object within a scene, and the AI automatically applies this to the entire video, maintaining the object’s position and shape even when the camera moves or objects are obscured. It also naturally renders lighting, shadows, and reflections to produce results that look as if they were actually filmed.
An image showing the character “S-Gom” from ZEM (Jam), SKTelecom’s kids-only brand, composited into a video on the SKTelecom YouTube channel using “Insert Anywhere.” Photo courtesy of SKTelecom
An image showing the character “Poxstar” from SKTelecom’s kids-only brand ZEM (Jam) composited into a video of a girl playing the guitar using “Insert Anywhere.” Photo courtesy of SKTelecom
SKTelecom has already commercialized this technology in “AdFlux,” its virtual product placement (VPPL) solution currently offered to advertising and media companies. The company has also filed a related patent application, and the technology has been applied to TV programs such as “Going All the Way: Dokbak Tour,” “Let’s Meet Now,” and “Laughter Out Loud—Call Anytime.”
Building on these research achievements, the company plans to expand the application of this technology to various industries requiring high-quality video compositing, such as video post-production, advertising production, and virtual product placement.
Yang Seung-ji, Head of Enterprise Technology Support at SKTelecom, stated, “The acceptance of this paper at ECCV 2026 signifies that SKTelecom’s AI video synthesis technology has been recognized for its world-class research competitiveness,” adding, “We will continue to advance our high-quality AI video synthesis technology so that it can be immediately applied in the advertising and media sectors.”
Consumers’ perspectives on brands are changing. There is a growing trend toward consuming not only a product’s taste or price but also the brand’s identity, atmosphere, and story. Particularly in the …
“Will you give me a name?”
A post titled “Newcomer Naming Service” featuring the official mascot, published on CJ Corp.’s official social media account. (Photo: Screenshot from CJ Corp.’s Instagram…
The Google app has surpassed Naver (NAVER(035420)) in monthly active users (MAU) in South Korea for three consecutive months. After the rankings flipped for the first time in July, Google maintained t…