Microsoft released a minute-long ad partially created with generative AI in January of this year. However, it was nearly three months before the company revealed that it had used AI to create it.
Image source: Microsoft
In a Microsoft Design blog post, senior design communications manager Jay Tan noted that the video generation process produced typical AI “hallucinations,” which required adjusting individual fragments and integrating them with the footage. When choosing scenes to generate, the team concluded that shots with complex motor skills, such as close-ups of hands typing on a keyboard, needed to be shot live. At the same time, short or static sequences were considered suitable for AI generation.
Microsoft did not specify which specific shots were generated by the AI, but Tan described the production process in detail. The team first used AI to create a script, storyboards, and pitch decks. Text queries and sample images were used to generate prompts, which were then fed into an image generator. The resulting images were edited and uploaded to Hailuo and Kling video generators. Other tools were not named.
According to creative director Cisco McCarthy, the team came up with thousands of different prompts, iterating on the results. “There’s never really a single, perfect prompt,” he said, so it was only through iteration that the quality was achieved. Visual designer Brian Townsend added that this approach cut up to 90 percent of the time and expense typically required for traditional video production.
Microsoft’s approach mirrors the position of the company’s head of design, Jon Friedman, who has previously said that AI is becoming a tool in the arsenal of creative professionals, not a replacement for them. According to him, the designer’s job today is not only to create, but also to edit, which is becoming increasingly important.
After Microsoft revealed that the video was AI-generated, the telltale signs of AI generation became apparent: an oversized glass jar, non-handwritten captions, and an overall visual design with a typical digital sheen. However, without knowledge of the AI, viewers would not have noticed these details for months. The editing, with frequent cuts, effectively smoothed out the visual artifacts that come with AI.