Quanta AI Labs

20 / May 12 – 18, 2025 · 10 RELEASES

AI News Week 20, 2025

Week 20 2025 AI releases.

Vision & Image (1)

  • Stitch

    Google · Text-to-Image

    Stitch is an AI-powered design tool that enables users to create visual content by leveraging generative AI capabilities integrated into a design interface. It streamlines the design workflow by combining traditional design tools with AI-assisted generation and editing features.

Video & Animation (2)

  • EX-4D

    Pico, Bytedance · Text-to-Video

    A novel framework for generating camera-controllable 360° videos from monocular input using Depth Watertight Mesh representation to handle extreme viewpoints and occlusions. Employs a lightweight LoRA-based video diffusion adapter trained on simulated multi-view data without requiring paired datasets.

  • LongAnimation

    USTC, HKUST · Video Editing

    A framework for long-term animation colorization using a Dynamic Global-Local Memory module that maintains color consistency across 500+ frame sequences by dynamically extracting and fusing global historical features with local generation features. Includes SketchDiT architecture, color consistency rewards, and supports text-guided creative background generation.

Agents & Automation (3)

  • Workflow Use

    browser-use · Web Agents

    A framework for building browser automation workflows using AI agents. Enables programmatic control of web browsers through natural language or structured task definitions.

  • AI Hedge Fund

    virattt · Agent Framework

    An open-source framework for building autonomous AI agents that perform financial analysis and portfolio management using multi-agent orchestration and LLM-based decision making.

  • Project Astra

    Google DeepMind · Embodied AI

    A multimodal AI agent that processes real-time video and audio inputs to understand and interact with physical environments, demonstrating reasoning and planning capabilities for embodied AI tasks.

Developer Tools (2)

  • Roboflow Trackers

    Roboflow · AI Infrastructure

    Open-source object tracking library providing multiple tracking algorithms for computer vision applications, enabling persistent identification of objects across video frames.

  • Gemini CLI

    Google · AI Infrastructure

    Open-source command-line interface for interacting with Google's Gemini AI models, enabling developers to build and deploy AI agents through terminal-based workflows.

Research & Safety (2)

  • AlphaGenome

    Google DeepMind · Research Paper

    AI system for genome analysis and understanding that applies deep learning to interpret genomic sequences and their functional implications. Advances protein structure prediction and genetic variant classification.

  • ShareGPT-4o-Image

    FreedomIntelligence · Dataset

    A dataset and framework for training multimodal models on image-text pairs, enabling GPT-4o-like image understanding capabilities. Provides shared resources for replicating advanced vision-language model training.

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