Today in Generative Media
Training AI models with public data; Google's $270M fine; Figma CEO's AI optimism
Here’s Proof You Can Train an AI Model Without Slurping Copyrighted Content (Wired)
Google hit with $270M fine in France as authority finds news publishers’ data was used for Gemini (TechCrunch)
Why Figma CEO Dylan Field is optimistic about AI and the future of design (The Verge)
Artist's guide to how to spot AI images is essential reading (CreativeBloq)
Rethinking How to Train Diffusion Models (Nvidia developer blog)
Character Voice For Everyone (Character.ai)
Write, Research, and Collaborate with your AI Personal Assistant (HyperWriteAI)
RadSplat: Radiance Field-Informed Gaussian Splatting for Robust Real-Time Rendering with 900+ FPS (project page)
HUGS: Holistic Urban 3D Scene Understanding via Gaussian Splatting (project page)
Wear-Any-Way: Manipulable Virtual Try-on via Sparse Correspondence Alignment (project page)
Magic Fixup: Streamlining Photo Editing by Watching Dynamic Videos (project page)
Mora is a multi-agent framework designed to facilitate generalist video generation tasks, leveraging a collaborative approach with multiple visual agents (GitHub) Paper on arXiv.
AnimateDiff-Lightning is a lightning-fast text-to-video generation model. It can generate videos more than ten times faster than the original AnimateDiff (HuggingFace) Demo.
We demonstrate semantic palette, a new drawing paradigm where users paint semantic meanings in addition to colors to create artworks. (GitHub) Demo on HuggingFace. Paper on arXiv.
Speedup Open-Sora's training by 3x and inference by 2x with our novel DSP (Dynamic Sequence Parallelism)! (X) Code on GitHub. Paper on arXiv.
IDAdapter: Learning Mixed Features for Tuning-Free Personalization of Text-to-Image Models (arXiv)