Today in Generative Media

Filmmaker's Sora notes; Half of US tried genAI; How to dodge Meta AI censorship

• I'm a filmmaker testing OpenAI's Sora. Here are its best features and biggest limitations — and the prompts I use. (Business Insider)

• More than half of U.S. has tried generative AI according to Adobe Analytics (VentureBeat)

• Meta AI's Keystroke Tracking Offers Loophole to Dodge Its Image Censors (PC Magazine)

• Adobe’s new Firefly model makes it easier to use Photoshop’s AI tools (The Verge)

• Create AI Influencers, Start Earning: Glambase lets your AI character take center stage, giving you the earnings and freedom you've always wanted. (glambase.app)

• LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation (HuggingFace)

• FlashFace: Human Image Personalization with High-fidelity Identity Preservation (Demo/API on Replicate)

• Improving Diffusion Models for Authentic Virtual Try-on in the Wild (project page) Demo/API on Replicate. My favorite part is that it works really well with out-of-the-distribution garments (X)

• Graphic Design with Large Multimodal Model (project page)

• PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generation (project page)

• Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis (project page)

• Analysis of Classifier-Free Guidance Weight Schedulers (arXiv)

• Contrastive Gaussian Clustering: Weakly Supervised 3D Scene Segmentation (arXiv)

• Does Gaussian Splatting need SFM Initialization? (arXiv)

• Video as the New Language for Real-World Decision Making (arXiv)

• VFX breakdown combining a variety of AI tools that shows the fluidity from pre-pro to post. (X)

• Here’s the thing: everyone in AI knows it will replace jobs. Even the most ardent gen AI proponents don’t claim that it won’t. And this is significant for a number of reasons. (X)