AI Gives Classic Films New Lease of Life
The 4K version of The Swordsman recently made its screening debut at the China Film Archive (the only national organization of film heritage preservation), commemorating the centenary of the birth of renowned novelist Louis Cha (JinYong) and the 20th anniversary of the passing of famous composer James Wong.
The debut is a part of the Classic Hong Kong Film Restoration Project. This initiative aims to preserve the legacy of Hong Kong cinema, once referred to as the "Hollywood of the East."
A collaboration between the China Film Archive, Douyin, and Volcengine (cloud service of Chinese tech company ByteDance), the project was launched in August 2023 to address the deterioration of film reels caused by aging. Over the past year, 100 films have been restored in 4K, with 90 utilizing AI-driven techniques and 10 combining AI with manual restoration.
AI technology has proven instrumental in streamlining the restoration process. Wang Zheng, a restoration specialist at the China Film Archive, explained that traditional methods, which involved manually removing scratches and stains, were both time-consuming and labor-intensive. AI now automates these repetitive tasks, significantly increasing efficiency. Intelligent frame interpolation technology, for example, enhances the fluidity of action sequences by doubling the frame rate.
Engineers from Volcengine also enhanced their visual AI models, addressing challenges such as limited detail in older films. Their advancements ensure that restored films meet modern 4K standards, achieving greater clarity and natural visual effects. Updates to AI algorithms have reduced restoration time by nearly threefold, with a two-hour film now requiring just five hours of processing instead of 18.
The project highlights the role of AI in cultural preservation, setting a precedent for its application in restoring and safeguarding cinematic heritage. Through this initiative, classic Hong Kong films are made accessible to contemporary audiences with enhanced visual fidelity.