Quanta AI Labs

30 / Jul 21 – 27, 2025 · 9 RELEASES

AI News Week 30, 2025

Week 30 2025 AI releases.

Language Models (2)

  • AI4Bharat Models

    AI4Bharat · Multilingual

    Collection of open-source AI models focused on Indian languages and applications. Platform provides access to multilingual NLP models optimized for Bharatian language support.

  • MiniCPM-V

    OpenBMB · Small Language Models

    A lightweight multimodal language model combining vision and language capabilities, designed for efficient deployment with reduced computational requirements compared to larger vision-language models.

Video & Animation (1)

  • OmniHuman-1.5

    ByteDance Intelligent Creation · Text-to-Video

    Generates expressive character animations from a single image and voice track using a dual-system architecture combining an MLLMs and Diffusion Transformer inspired by cognitive theory. Supports rhythmic performances, emotional acting, multi-person scenes, and text-guided control with audio synchronization across diverse character types.

Audio & Speech (1)

  • OpenVoice

    MyShell AI · Voice Cloning

    Open-source instant voice cloning tool that replicates speaker characteristics across languages with minimal audio samples. Enables zero-shot voice style transfer and multilingual synthesis without retraining.

3D & Spatial (3)

  • HandMDM

    LIGM École des Ponts, NVIDIA · 3D Generation

    A text-conditioned diffusion model for generating 3D hand motions from natural language descriptions, trained on large-scale sign language video data with LLM-generated motion annotations. The model generalizes across unseen sign categories, different sign languages, and non-sign hand movements.

  • VoxHammer

    Beihang University, Renmin University, Tsinghua University, Tencent · 3D Reconstruction

    Training-free 3D editing method that performs precise and coherent editing in native 3D latent space by inverting 3D models and selectively replacing denoising features of preserved regions to maintain consistency while integrating edits.

  • Pixie: 3D Physics from Pixels

    University of Pennsylvania & MIT · PBR Materials

    Pixie is a neural network that maps pretrained visual features to dense material fields (Young's modulus, Poisson's ratio, density) in a single forward pass, enabling real-time physics simulations from images. It achieves 1.46-4.39x better realism than existing methods while running 1000x faster, trained on PixieVerse, a large-scale synthetic dataset with automatic VLM-based annotations.

Agents & Automation (2)

  • Agentic Doc

    Landing AI · Agent Framework

    A framework for building document-understanding agents that can process, analyze, and extract information from documents using agentic AI patterns. Enables autonomous document workflow automation with multi-step reasoning capabilities.

  • Mem0

    Mem0 AI · Agent Framework

    Open-source memory layer for AI applications that enables persistent, personalized context management across conversations and sessions. Provides a unified API for adding, retrieving, and managing memory for LLMs and agents.

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