Today in Generative Media

AI terms cut from actors union offer; Nintendo denies AI use; Leo XIV thinks about AI

News and Opinion

• Video Game Companies Remove AI Provision on Unlimited Digital Replicas From ‘Best, Last and Final’ Offer to Actors Union (EXCLUSIVE) (Variety)

• Nintendo: No, We Didn't Use AI for Mario Kart World Artwork (PC Magazine)

• Nintendo Insists AI-Generated Images Were Not Used in the Development of Mario Kart World After In-Game Billboards Sparked Speculation (IGN)

• Pope Leo XIV says advancement of AI played a factor in his papal name selection (NBC News)

• I scammed my bank. All it took was an AI voice generator and a phone call (Business Insider)

• Tall Tales is a critique of AI — so why do people think it was made with AI? (The Verge)

• I'm a filmmaker who used AI to make a sci-fi film for under $300. I'm deeply conflicted about it. (Yahoo! Entertainment)

• Going ‘AI first’ appears to be backfiring on Klarna and Duolingo (Fast Company)

• Google to unveil Imagen 4, Imagen 4 Ultra and Veo 3 models at I/O 2025 (Testing Catalog)

Software

• [Luma] Reframe is here. Outpaint and resize any uploaded video, image, or DreamMachine creation into every format you need. (X)

• 🌟 New Feature Alert 🌟 HeyGen Avatar IV is here! (HeyGen)

• Introducing Meta 3D AssetGen 2.0: A new foundation model for 3D content creation (Meta Horizon blog)

• ZenCtrl Demo One framework to Generate multi-view, diverse-scene, and task-specific high-resolution images from a single subject image—without fine-tuning. (HuggingFace)

• …the most practical UX for GPT-4o image creation. Drag & drop references. Re-use prompts and parameters. One-click suggestions, "prompt rewrites“, translations & and more. (X)

• Amazon Nova Sonic State-of-the-art speech-to-speech model for conversational AI

• Ace-Step Audio Model Native Support in ComfyUI! (ComfyUI Blog)

• Gemini 2.5 Models now support implicit caching (Google Developers blog)

Research

• Pixel3DMM: Versatile Screen-Space Priors for Single-Image 3D Face Reconstruction (project page)

• HunyuanCustom: A Multimodal-Driven Architecture for Customized Video Generation (project page)

• CompleteMe: Reference-based Human Image Completion (project page)

• PixelHacker: Image Inpainting with Structural and Semantic Consistency (project page)

• GarmentDiffusion: 3D Garment Sewing Pattern Generation with Multimodal Diffusion Transformers (project page)

• KeySync: A Robust Approach for Leakage-free Lip Synchronization in High Resolution (project page)

• Generating Animated Layouts as Structured Text Representations (project page)

• Introducing Multiverse: The First AI Multiplayer World Model (project page)

• AnimateAnywhere: Rouse the Background in Human Image Animation (project page)

• S3D: Sketch-Driven 3D Model Generation (arXiv)

Misc

• Flow Matching Guide and Code (arXiv)

• The Little Book of Deep Learning [full PDF, 185pp.] (Francois Fleuret)

• Testing the Insta360 X5 for gaussian splatting. (X)

• Token & Trinket is a game where one model plays the shopkeeper (in this case Sonnet 3.7), managing store finances, purchasing, inventory, pricing, debt payments etc. Another model (Opus) plays gamemaster, choosing from a set of events that affects the economics and building a narrative around it. (X)