This weekend in Generative Media

YouTube's audio eraser; OpenAI asks NYT to prove authorship; AI's telltale words

News

• YouTube’s updated eraser tool removes copyrighted music without impacting other audio (TechCrunch)

• ChatGPT's Owner Asks For Proof That 'New York Times' Articles Are Original. (Entrepreneur)

• The telltale words that could identify generative AI text (Ars Technica)

• UK Says No To AI. Virtual Candidate Finishes Last In Polls (NDTV World)

• OpenAI breach is a reminder that AI companies are treasure troves for hackers (TechCrunch)

Software

• AI braintrust for creative teams (Promethean AI)

• Portrait animation using a driving video source (LivePortrait API on Replicate)

• We recognize that the commercial license originally associated with SD3 caused some confusion and concern in the community so we have revised the license for individual creators and small businesses. Models released under this new “Stability AI Community License” can be used for free much more broadly than they could under the previous licenses. (Stability)

Research

• CharacterGen: Efficient 3D Character Generation from Single Images with Multi-View Pose Calibration (SIGGRAPH 2024, project page)

• ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024, project page)

• E.T. the Exceptional Trajectories: Text-to-camera-trajectory generation with character awareness (ECCV 2024, project page)

• LaRa: Efficient Large-Baseline Radiance Fields (ECCV 2024, project page)

• HouseCrafter: Lifting Floorplans to 3D Scenes with 2D Diffusion Models (project page)

• SVG: 3D Stereoscopic Video Generation via Denoising Frame Matrix (project page)

• Segment Any 4D Gaussians (project page)

• fVDB: A Deep-Learning Framework for Sparse, Large-Scale, and High-Performance Spatial Intelligence (arXiv)

• Diffusion Models and Representation Learning: A Survey (arXiv)

Misc

• From a recent Goldman Sachs report on generative AI: limited $$ upside, capital intensive, can't solve complex problems, no killer app. The bull case is that they somehow figure it all out or simply that "bubbles take a long time to burst." (X)

• Something gen-AI-3D people really do not grasp is that the tools & workflows available to “clean up” issues in generated 3D content are so, so painful and tedious compared to the create-from-scratch tools. It’s not remotely like the situation in image & video editing (X)

• The next generation of vtubers might look like this. (LinkedIn)

• This was done in less than 24h by one person using AI as the ground tooling, some post in AE and that’s it. (X)

• I think real time video gen at current quality sometime next year for sure real time video gen on B200s or grok etc and in 2026/2027 we have Hollywood-level movies fully generated (not real time) leveraging existing media data. (X)

• Kling image to video, Kraken attacking a ship at sea. Sound created in Elevenlabs. (X)

• A man reacts to rocks, a chef puts far too much salt on some food (nice salt physics), hard math is explained, a woman reacts to a potato, a Twitch streamer, & flaming car review (X)