This weekend in Generative Media

GenAI book scams; Sora shouldn't rattle Hollywood; Meta's image AI flunks history too

• Tech writer Kara Swisher has a new book. Enter the AI-generated scams. (Washington Post)

• OpenAI’s Sora isn’t even here yet, and it’s already rattling Hollywood. Maybe it shouldn’t be. (FastCompany)

• Meta AI creates ahistorical images, like Google Gemini (Axios)

• An update on the BBC’s plans for Generative AI (Gen AI) and how we plan to use AI tools responsibly (BBC)

• SAG-AFTRA Executive Committee Approves TV Animation Agreement, Sends To Members For Ratification Vote (Deadline) “The new contract includes language denoting voice actors as ‘only humans’ and also established regular, mandatory artificial intelligence meetings with producers, which will include discussion of methods and systems to track the use of digital replicas.”

• Lou Reed’s widow resurrected him as an AI, “sadly addicted” to talking to it. (Futurism)

• Today, we are thrilled to launch a project called Open-Sora plan, aiming to reproduce OpenAI's video generation model, and have received unanimous anticipation from netizens both domestically and internationally. (project page)

• Trajectory Consistency Distillation (project page)

• Splatter Image: Ultra-Fast Single-View 3D Reconstruction (project page)

• ViewFusion: Towards Multi-View Consistency via Interpolated Denoising (project page)

• SongComposer: A Large Language Model for Lyric and Melody Composition in Song Generation (project page)

• ChatMusician: Understanding and Generating Music Intrinsically with LLM (project page)

• Introducing SwitchLight: Be anywhere at any time. (X) Paper on arXiv. Waitlist on Beeble.

• StyLitGAN, a simple approach for image-based relighting without labeled or paired data, has been accepted to CVPR2024. The first version of it was submitted to NeurIPS 2022, and after that rejection, it changed my view on generative models. A🧵on our persistent journey: (X)