Today in Generative Media
Disney, Universal sue Midjourney; Anthropic's AI blog ends; China stops AI for exams
News and Opinion
Disney and Universal Sue A.I. Firm for Copyright Infringement (New York Times) The suit, against Midjourney, is the first time major Hollywood companies have sued over A.I.-generated images.
Anthropic’s AI-generated blog dies an early death (TechCrunch)
China shuts down AI tools during nationwide college exams (The Verge)
Ohio State University says all students will be required to train in AI (The Guardian)
Like humans, AI is forcing institutions to rethink their purpose (VentureBeat)
X’s new policy prevents companies from using posts to ‘fine-tune or train’ AI models (The Verge)
AI Used To "Extend The Artist" Is Very Different Than "AI Art" (And That Gives Me Hope) (Peter Csathy on Substack)
Software
Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models (CVPR 2025) (GitHub)
Comma v0.1 1T and 2T—7B LLMs trained on openly licensed text (Simon Willison’s blog) MLX is still a very new format, but Claude 4 Sonnet has a training cutoff date of March 2025 so I crossed my fingers and hoped it would be able to help me out. It did exactly that!
Looks like we have a new best open-weights (Apache 2) embedding model from Qwen (X)
Research
OmniSync: Towards Universal Lip Synchronization via Diffusion Transformers (project page)
Exploring Timeline Control for Facial Motion Generation (project page)
Cora: Correspondence-aware image editing using few step diffusion (project page)
RaySt3R: Predicting Novel Depth Maps for Zero-Shot Object Completion (project page)
Follow-Your-Creation: Empowering 4D Creation through Video Inpainting (project page)
Removing Reflections from RAW Photos (CVPR open access)
UMA: Ultra-detailed Human Avatars via Multi-level Surface Alignment (arXiv)
HuGeDiff: 3D Human Generation via Diffusion with Gaussian Splatting (arXiv)
Misc
Reverse Engineering Cursor's LLM Client (TensorZero blog)