爱奇艺用 NadouPro AI 工作流重建 1938 年失传影片

An eight-person team used iQIYI’s NadouPro platform to reconstruct a 1938 film whose prints, reels and scripts had been lost, the company announced on November 6, 2026. The short, completed in two months, was shortlisted for the 2026 Future Images AI Unit at the 39th Golden Rooster Awards.

The team worked at the iQIYI-XMU Joint Laboratory of Film Restoration. Its task differed from conventional restoration, which uses digital technology to repair damage and enhance image quality in surviving material. Visiting Shanghai after Sixty Years had been lost for nearly 90 years, leaving production still photographs, posters and newspaper clippings as the conceptual guides for generating a new version.

iQIYI describes the original as one of China’s earliest science-fiction works, a comedy that imagined life sixty years into the future.

Zidong Jiang, a joint-laboratory member and iQIYI’s director of AI restoration technology, said the team therefore pursued a creative reconstruction based on available historical materials rather than attempting to restore a surviving print.

Fixed Digital Assets Address Continuity

The production highlighted two central difficulties in generative video: maintaining character consistency across scenes and achieving precise camera control. The story spans 1938 and 1998 and features multiple characters, requiring continuity across eras as well as individual shots.

NadouPro served as a centralized production hub. The team fixed digital assets for each character’s facial features, proportions and costumes before generating subsequent shots from those foundational files, rather than creating each character anew for every scene. iQIYI says this process kept character appearances consistent throughout the film.

Recreating performances from still photographs required detailed control over the generated images. NadouPro’s 3D Director Studio allowed the team to adjust character positioning, composition and movement, with additional tools for testing scenes, visual styles and camera angles.

Director Enji Yang credited the platform with helping the team move from relatively random image generation toward shots with clear directorial intent. The early production work still followed conventional steps: script analysis, character design, scene design and storyboard planning. “We always believe the audiovisual design of a film must come first, before considering how AI can realize it,” Yang said.

The filmmakers gave the two eras distinct visual treatments. The 1938 sequences drew on early black-and-white cinema through stable camerawork and vintage lighting, while the 1998 scenes adopted natural light and handheld camera movement in a documentary style.

iQIYI has explored AI film restoration since 2017. Its proprietary ZoomAI uses algorithms for scratch repair, super-resolution, noise reduction and intelligent frame interpolation. The company says these capabilities help shift restoration from manual frame-by-frame work toward automated processes. Through its joint laboratory with Xiamen University, iQIYI has previously brought a number of classics back to screens in 4K.

Aigong Zhang, the short’s chief planner and director of the joint laboratory, described the project as an integration of traditional filmmaking craftsmanship with AI. That combination is visible in a workflow where historical records and conventional preproduction guide the creative choices, while NadouPro supplies tools for generating and controlling the images.

The project extends iQIYI’s restoration work into a different category. Fixed character assets and explicit directorial controls address practical production problems, but the surviving archive still defines what can be reconstructed. The resulting short is a new interpretation guided by historical evidence; it does not recover the lost original footage.