Quanta AI Labs

37 / Sep 8 – 14, 2025 · 9 RELEASES

AI News Week 37, 2025

Week 37 2025 AI releases.

Language Models (1)

  • MiniMax-M1

    MiniMax-AI · Small Language Models

    MiniMax-M1 is a compact language model optimized for efficient inference and deployment. The model targets low-latency performance on resource-constrained environments.

Vision & Image (1)

  • TAIR: Text-Aware Image Restoration with Diffusion Models

    KAIST, Korea University, Yonsei University, Samsung Electronics · Image Restoration

    Introduces Text-Aware Image Restoration (TAIR), a diffusion-based framework that recovers both visual quality and text fidelity in degraded images using a multi-task approach with text-spotting features. Presents SA-Text, a 100K image benchmark with dense text annotations, to address the text hallucination problem in existing image restoration methods.

Video & Animation (1)

  • PlayerOne

    HKU, DAMO Academy (Alibaba), Hupan Lab, HUST · Text-to-Video

    First egocentric world simulator that generates realistic first-person videos from a single scene image and human motion sequences, using a coarse-to-fine training pipeline with part-disentangled motion injection and joint 4D scene-video reconstruction for world-consistent long-form generation.

3D & Spatial (3)

  • PartCrafter

    Peking University, ByteDance PICO, Carnegie Mellon University · 3D Generation

    PartCrafter is a structured 3D generative model that jointly synthesizes multiple semantically meaningful 3D meshes from a single RGB image using compositional latent diffusion transformers with hierarchical attention mechanisms. The method generates part-decomposable objects and multi-object scenes end-to-end without requiring pre-segmented inputs.

  • Hunyuan3D-2.1

    Tencent · 3D Generation

    3D generation model from Tencent that creates 3D assets from text or image inputs. Improves upon previous version with enhanced quality and generation speed for 3D content creation.

  • ImmerseGen

    PICO, Bytedance & Zhejiang University · 3D Generation

    ImmerseGen generates panoramic 3D VR worlds from text prompts using VLM-based agents to design and arrange alpha-textured proxy assets, combined with terrain synthesis and ambient sound generation. The method creates immersive environments with compact assets while maintaining diversity and realism without requiring complex pre-built asset libraries.

Developer Tools (1)

  • Awesome MCP Clients

    punkpeye · AI Infrastructure

    A curated collection of Model Context Protocol (MCP) clients and implementations, serving as a resource hub for developers building with the MCP standard. Aggregates existing MCP client tools and frameworks across different platforms and languages.

Research & Safety (2)

  • Weather Lab Cyclone Predictions

    Google DeepMind · Research Paper

    AI system for improved tropical cyclone prediction and tracking using machine learning models trained on historical weather data. Demonstrates application of neural networks to enhance meteorological forecasting accuracy.

  • LoRAEdit

    cjeen (GitHub) · Research Paper

    A method for editing neural networks using Low-Rank Adaptation (LoRA) to enable parameter-efficient model modifications. Enables targeted fine-tuning and adaptation of pretrained models through low-rank decomposition.

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