China’s Kuaishou debuts multimodal AI video model with ‘Nano Banana’ editing capability

Chinese short video platform Kuaishou on Monday launched a new artificial intelligence video generation model to challenge OpenAI’s Sora and start-up Runway in the global market for AI content creation.\nKuaishou, which competes with TikTok’s Chinese sibling Douyin in China, said Kling O1 was the first unified multimodal video model in the industry, based on an architecture that integrated diverse video creation tasks – generation, precise and controllable editing and understanding – into a single platform, providing a “seamless, end-to-end workflow for the creative industry”, according to a statement.\nKling O1 was hailed as the “Nano Banana for AI video”, according to Alvaro Cintas-Canto, an assistant professor of AI and cybersecurity at Marymount University in the US. He lauded the video tool’s versatility in handling text-to-video, content editing and maintaining video character consistency in complex scenes in a post published to his X account on Monday.\nNano Banana is Google’s image generation and editing model, which is known for its precise manipulation of visual elements in a photo. Kuaishou claimed to be the first in the industry to build the same advanced editing capabilities into a video generation model, paving the way for its use in real-world settings across industries.\n\nKuaishou credited Kling O1’s multimodal capabilities and semantic comprehension for allowing it to take in and understand different content formats from text, images, video and visual elements. This lent it the ability to “see and understand” different parts and perspectives of an image, video or character that users upload as a reference when generating videos, while retouching video content in a more precise fashion.\nKling O1 supports a range of complex editing functions, such as swapping in characters, editing out backgrounds, retouching partial video content while keeping everything else intact and fine-tuning other details.\nKuaishou expected the new model’s abilities to drive its adoption by filmmakers, production studios, advertisers and influencers who want to streamline their workflows.\nIn the domestic market, Kuaishou is up against Chinese video tool makers from ByteDance to Tencent Holdings, while it competes with OpenAI, Google and Runway in the international market.\nIts Kling 2.5 Turbo model, released in September, was the world’s best for image-to-video generation and ranked No 3 for text-to-video, trailing only Runway and Google’s Veo 3, according to AI analytics firm Artificial Analysis.\nKuaishou has demonstrated its Kling AI models at prestigious global film festivals, including Cannes in May and Tokyo in October, showcasing how the film industry could benefit from advances in AI tools.\nCheng Yixiao, Kuaishou co-founder and chief executive, said on a recent earnings call that the company’s AI strategy would focus on using the technology to aid in TV and film content creation.\nThe company’s Kling AI business, which charges users a premium for access to its video tools, generated 300 million yuan (US$42 million) in sales in the third quarter, according to its latest financial results.\n