This weekend in Generative Media

The first AI beauty pageant; Shutterstock's $104M AI deals; Perplexity rips off news

News

• The Uncanny Rise of the World’s First AI Beauty Pageant (Wired)

• Shutterstock’s AI-Licensing Business Generated $104 Million Last Year (Bloomberg)

• Buzzy AI Search Engine Perplexity Is Directly Ripping Off Content From News Outlets (Forbes)

• Researchers Use AI to Decode the Secret Language of Dog Barks (Gizmodo)

• Google's note-taking AI app can now read from the web (NotebookCheck)

Products

• PixVerse Magic Brush (PixVerse)

• Meet Ultravox, our open source multimodal LLM. Check out our v0.1 release at https://ultravox.ai (X)

• RenderNet has just introduced CharacterLab, a new concept in character creation. Imagine creating consistent characters in less than a minute and having them always available. (X)

• Freepik's online editor is finally here and its name is 'Freepik Designer' (X)

Code

• Stable-Pose is a novel adapter that leverages vision transformers with a coarse-to-fine pose-masked self-attention strategy, specifically designed to efficiently manage precise pose controls during Text-to-Image (T2I) generation. (GitHub)

• BIRD is a new image restoration method and can restore images from Gaussian blur, motion blur, and JPEG compression artifacts in just a few seconds (X) Code on GitHub.

Research

• GaussianPrediction: Dynamic 3D Gaussian Prediction for Motion Extrapolation and Free View Synthesis (SIGGRAPH 2024, project page)

• MotionFollower: Editing Video Motion via Lightweight Score-Guided Diffusion (project page)

• Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image (project page)

• Fourier123: One Image to High-Quality 3D Object Generation with Hybrid Fourier Score Distillation (project page)

• Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image (project page)

• WE-GS: An In-the-wild Efficient 3D Gaussian Representation for Unconstrained Photo Collections (project page)

• Freeplane: Unlocking Free Lunch in Triplane-Based Sparse-View Reconstruction Models (project page)

• Generalizable Human Gaussians from Single-View Image (project page)

• Coarse-To-Fine Tensor Trains for Compact Visual Representations (project page)

• 4Diffusion: Multi-view Video Diffusion Model for 4D Generation (project page)

• PR3D: Precise and realistic 3D face reconstruction from a single image (Computer Animation and Virtual Worlds)

• De-NeRF: Ultra-high-definition NeRF with deformable net alignment (Computer Animation and Virtual Worlds)

• A Simple Approach to Differentiable Rendering of SDFs (arXiv)

• Localize, Understand, Collaborate: Semantic-Aware Dragging via Intention Reasoner (arXiv)

Misc

• the fact that you can generate a nice HD image in 3-4 seconds and then create a 3D model from it in ~60 seconds on Hugging Face (for free) is 🔥 (X)

• China has a huge advantage in the AI race: total disregard for IP laws. They will just train huge models on every movie, every song, all of YouTube (X)

• The biggest unlock of what new frontier models will do is less tied to things we weren't able to do before and much more related to the dramatic expansion of the number of creative minds that will not be constrained by tool rendering time and resources. (X)