28 / Jul 7 – 13, 2025 · 12 RELEASES

AI News Week 28, 2025

Week 28 2025 AI releases.

Language Models (1)

UQLM
CVS Health · Text Generation
Utility-based Query Language Model designed for healthcare applications. A specialized LLM framework for processing and understanding medical queries and documentation.

Stand-In
WeChatCV · Visual Understanding
A computer vision project from WeChat's CV team for pose estimation and human synthesis tasks. Enables realistic human figure replacement and animation in images and videos.
Voost
NXN Labs (Seungyong Lee, Jeong-gi Kwak) · Image Editing
A unified diffusion transformer framework that jointly learns virtual try-on and try-off by modeling both tasks bidirectionally, achieving state-of-the-art results through shared garment-body relational reasoning without task-specific networks. Introduces attention temperature scaling and self-corrective sampling techniques to improve robustness and leverage bidirectional consistency.

Matrix-Game 2.0
Skywork AI · Real-time Video
An open-source interactive world model that generates long videos in real-time (25 FPS) using few-step auto-regressive diffusion, with frame-level action injection for mouse/keyboard control across game environments like Minecraft, GTA5, and Unreal Engine.
Yan
Yan Team · Text-to-Video
Foundational framework for interactive video generation comprising three modules: Yan-Sim for high-fidelity 1080p 60fps game simulation using causal diffusion, Yan-Gen for multimodal text/image-to-interactive-video generation via DiT backbone, and Yan-Edit for multi-granularity real-time editing via hybrid mechanics simulator and visual renderer.
FantasyPortrait
Alibaba AMAP & Beijing University of Posts and Telecommunications · Text-to-Video
A diffusion transformer framework for generating high-fidelity multi-character portrait animations with expression-augmented learning and masked cross-attention to prevent feature interference. Supports video-driven and audio-driven animation across multiple languages with minimal fine-tuning.
StableAvatar
Fudan University, Microsoft Research Asia, Xi'an Jiaotong University, Tencent · Text-to-Video
End-to-end video diffusion transformer that generates infinite-length audio-driven avatar videos without post-processing, using a novel Time-step-aware Audio Adapter to prevent latent distribution error accumulation and an Audio Native Guidance Mechanism for enhanced synchronization.

Matrix-3D
Skywork AI · 3D Generation
Framework for omnidirectional explorable 3D world generation from images or text using trajectory-guided panoramic video diffusion and feed-forward/optimization-based 3D reconstruction. Introduces Matrix-Pano dataset with 116,759 synthetic panoramic video sequences with precise camera annotations.
VertexRegen
Meta Reality Labs Research / UC San Diego · 3D Generation
A mesh generation framework that enables coarse-to-fine generation with continuous level of detail by reversing edge collapse operations (vertex splits) through a learned generative model. Produces comparable quality to state-of-the-art methods while allowing anytime generation and halting at any step for valid meshes of varying detail levels.

Hunyuan GameCraft
Tencent · AI Infrastructure
AI system for game content generation and design, enabling automated creation of game assets, levels, and mechanics through multimodal AI capabilities.
ComfyUI-Copilot
AIDC-AI · AI Infrastructure
An AI assistant extension for ComfyUI that provides intelligent guidance and automation for node-based image generation workflows. Enhances the ComfyUI interface with copilot capabilities to streamline creative tasks.
Firecrawl
Firecrawl Team · Web Scraping
Open-source web scraping and data extraction tool that converts websites into clean, structured markdown and JSON for LLM consumption. Provides API and self-hosted options for automated web data processing.

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.