Today in Generative Media

Adobe previews video genAI; Apple's new AI is mid; 3rd Gen:48 AI film competition

News

• Adobe previews its upcoming text-to-video generative AI tools (The Verge)

• Apple’s new AI is magically mediocre (Vox)

• Gen:48 is a short film competition where teams of 1-3 filmmakers have 48 hours to ideate and execute a 1-4 minute film. This is the third edition and will take place from 9AM on September 14th until 9AM, September 16th (EST). (RunwayML)

• China refuses to sign agreement to ban AI from controlling nuclear weapons (Yahoo)

Software

• Meta Llama: Everything you need to know about the open generative AI model (TechCrunch)

• We just shipped a new variant of Structured Outputs in the Gemini API called Enum Mode, which allows you to easily constrain the model to pick between pre-defined options (X)

• We are in the process of relaxing some of the restrictions Colab has long had around webui usage. You should find Gradio & other webuis a little easier to use (X)

• DeepSeek-V2.5 wins praise as the new, true open source AI model leader (VentureBeat)

• PuzzleAvatar: Assembly of Avatar from Unconstrained Photo Collections (GitHub)

• xDiT is an inference engine designed for the large scale parallel deployment of Diffusion Transformers (DiTs) (HuggingFace)

• OPEN-MAGVIT2: An Open-source Project Toward Democratizing Auto-Regressive Visual Generation (GitHub)

Research

• SongCreator: Lyrics-based Universal Song Generation (project page)

• LinFusion: 1 GPU, 1 Minute, 16K Image (arXiv)

Misc

• Runway Gen3 - Quick scifi tests in Turbo mode (X)