New AI Video Goes Hard At Open AI!

The video script introduces a new AI video generator called 'Vu', which is being compared to Sora, a yet-to-be-released model. The presenter acknowledges the irony of comparing it to Sora before its launch. The video dives into the features of the new model, its potential to match Sora's quality, and the possibility of its use before Sora's release. The script also mentions a signup link for the audience. Vu is developed by Shinu Technology and Singua University, and it targets creating 16-second clips at 1080p resolution. The architecture of Vu is based on the Universal Video Transformer (UViT), which is a combination of two research papers: DPM solver for better predictions in diffusion models and 'All Are Worth Words' for combining Vision Transformers with a Unet model. Vu's strength lies in its ability to treat all elements as tokens and utilize long skip connections for coherent video generation.

The script provides an analysis of full 16-second clips generated by the Vidu AI, highlighting the references to Sora in the initial hype reel. It discusses the temporal coherence of the generated content, comparing it to Sora's outputs. The presenter appreciates the mid-journey V4 aesthetic of the TVs in one of the clips, which is reminiscent of a favorite model. Another clip features a panda bear playing a guitar, which, while not the most realistic, still impresses with its background coherence and reactive shadow. A beach vacation villa clip showcases an interesting dissolve between shots, hinting at the model's ability to handle transitions. An imaginative clip with a ship in a bedroom demonstrates the model's reaction to movement and environmental interaction. The script also includes a brief comparison with Sora, noting that while Sora's environment realism is slightly superior, Vidu's output still appears as a real place. The presenter reminds the audience that both models have their strengths and that the examples shown are cherry-picked, with Sora also producing less consistent videos that require significant post-production work.

The video script concludes with a discussion on the post-production process necessary to refine AI-generated videos into a final product. It mentions the use of AI tools in creating compelling imagery and the effort that goes into making these videos look semi-consistent. The presenter references a production company's use of Sora to create a short film, 'Airhead', and the extensive cleanup required to achieve a polished result. The script also highlights the creative process used by Paul Trello in his short film 'Notes to My Future Self', where AI imagery was integrated with traditional VFX techniques. Finally, the presenter provides a signup link for Vidu, noting a potential temporary issue with the website's submit button, and teases an upcoming interview with Adobe about Sora's integration into Premiere and future plans for After Effects.