Descriptions:
This AI news roundup from AI Search covers the most significant open-source releases from the final week of December 2025. The video leads with Nvidia’s Nitrogen, a vision-action foundation model trained on 40,000 hours of gameplay across more than 1,000 titles. Rather than accessing game code directly, Nitrogen plays by observing the screen and issuing controller inputs — and can generalize to games it was never trained on, handling multi-step tasks like vehicle flips and boss combat.
Alibaba’s Tongyi Lab contributes Flashportrait, an infinite-length portrait animation system claimed to run six times faster than comparable tools including Live Portrait, Fantasy Portrait, and Hunyuan Portrait, while maintaining character consistency across videos longer than one minute. The roundup also covers a new open-source image editor with natural language editing capabilities similar to Adobe Firefly, an AI photo-refocusing tool, a ByteDance Storymen release, and a 3D scene reconstruction system (requiring 50+ GB of VRAM) that breaks scenes into individually editable objects.
Two new open-source language models are highlighted as rivaling Gemini 3 and GPT 5.2 in performance. A Unitree humanoid robot teleoperation demo — showing full-body mirroring without bulky motion-capture equipment — closes out the video. All featured projects link to public GitHub repositories with local setup instructions, making this a dense reference for tracking the open-source AI frontier heading into 2026.
📺 Source: AI Search · Published December 28, 2025
🏷️ Format: Roundup
