Quanta AI Labs

19 / May 5 – 11, 2025 · 12 RELEASES

AI News Week 19, 2025

Week 19 2025 AI releases.

Language Models (4)

  • Hunyuan

    Tencent · Text Generation

    Tencent's multimodal AI model family supporting text generation, image generation, and video generation capabilities. Designed for enterprise and consumer applications with multilingual support.

  • Wan2.1-VACE-14B-GGUF

    QuantStack · Text Generation

    A 14B parameter language model in GGUF format optimized for quantized inference. Wan2.1-VACE appears to be a specialized variant designed for efficient deployment on consumer hardware.

  • Codex

    OpenAI · Code Generation

    A large language model trained on public code that can translate natural language into executable code across multiple programming languages.

  • Codex

    OpenAI · Code Generation

    A GPT-based model trained on code that can interpret natural language and generate functional code across multiple programming languages. Accessible via API for translating English descriptions into executable programs.

Vision & Image (2)

  • LightLab

    Google Research / Tel Aviv University / Hebrew University · Image Editing

    Diffusion-based method for parametric control over light sources in single images, enabling adjustment of light intensity/color, ambient lighting, and virtual light insertion while simulating complex effects like shadows and reflections through a combination of real photograph pairs and synthetic training data.

  • BLIP-3o

    Jiuhai Chen · Visual Understanding

    A multimodal vision-language model extending BLIP architecture with unified vision-language understanding across images and text. Appears to focus on omni-directional visual reasoning capabilities.

Video & Animation (2)

  • LTX-Video

    Lightricks · Text-to-Video

    A video generation model capable of creating high-quality video content from text or image prompts. Designed for efficient inference and integrated into Lightricks' creative platform.

  • VACE

    Tongyi Lab (Alibaba) · Video Editing

    All-in-one video generation and editing model supporting 6+ capabilities (Move-Anything, Swap-Anything, Reference-Anything, Expand-Anything, Animate-Anything) with unified control over content, structure, subject, posture, and motion preservation through multiple conditioning modalities (masks, depth, pose, trajectories, references).

Audio & Speech (2)

  • Stable Audio Open Small

    Stability AI & ARM · Text-to-Speech

    Lightweight audio generation model optimized for on-device deployment, enabling real-time audio synthesis and control with reduced computational requirements through ARM collaboration.

  • Stable Audio Tools

    Stability AI · Audio Processing

    Open-source toolkit for generating and manipulating audio using diffusion models, enabling text-to-audio synthesis and audio editing capabilities.

Agents & Automation (1)

  • DeerFlow 2.0

    DeerFlow · Agent Framework

    Open-source super agent framework enabling complex multi-step tasks through sandboxed execution, long/short-term memory, sub-tasking, and extensible skills with support for multiple LLM providers. Evolved from deep research agent to full-stack autonomous system with persistent file system, Docker-based isolation, and flexible tool composition.

Developer Tools (1)

  • n8n

    n8n · AI Infrastructure

    Open-source workflow automation platform that enables visual orchestration of APIs and services without code. Provides node-based interface for building complex integrations and automation sequences.

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