Alibaba Group Holding Ltd. released Wan 2.1, a new open-source AI video generation model on February 25th, which is now available on platforms such as ChatLLM, Replicate, and KREA AI. The 14b model supports text-to-video and image-to-video conversion, generating videos in resolutions up to 720p. It is said to outperform competitors like Google's Veo-2 and OpenAI's Sora in quality, while offering less censorship. Wan 2.1 has been praised for its ability to generate videos with complex motions, realistic physics, and cinematic quality. A 5-second video at 480p resolution takes about 39 seconds to generate, while at 720p it takes around 150 seconds. The model also supports the generation of text within videos in both Chinese and English, along with sound effects and background music that align with the visual content. The model's open-source nature allows for widespread use and modification, with versions available that can run on consumer-grade hardware. However, within 24 hours of its release, Wan 2.1 has been adopted by communities creating AI-generated pornography, raising concerns about its potential misuse and turning it into an AI pron machine.
A very quick test of Wan 2.1 on @krea_ai . I gave it a Midjourney image and the text prompt "Confusion", which was the theme of the original image. Wan 2.1 seems to run at a low frame rate, which is frustrating, but it is pretty creative. It constructed the front of the woman… https://t.co/ByJdeSVZJq
awesome & powerful the new ai video model Wan 2.1 next level via @krea_ai https://t.co/yUgEenh1Ep
Text-to-video AI showdown! 🎬 We put Google's Veo 2 against China's new Wan 2.1 using identical prompts to see which creates more realistic videos. Both models impressed us with their understanding of realism, each with unique strengths and weaknesses. After careful comparison,… https://t.co/4t1sAIzHje