Quanta AI Labs

12 / Mar 17 – 23, 2025 · 9 RELEASES

AI News Week 12, 2025

Week 12 2025 AI releases.

Language Models (1)

  • Lumina-mGPT-2.0

    Alpha-VLLM · Text Generation

    A multimodal language model iteration that appears to advance vision-language capabilities. Specific technical improvements unavailable due to truncated page content.

Vision & Image (3)

  • EasyControl

    Tiamat AI, ShanghaiTech University, National University of Singapore, Liblib AI · Text-to-Image

    EasyControl is a framework for efficient multi-condition controlled image generation using Diffusion Transformers, featuring a lightweight LoRA-based condition injection module, position-aware training for flexible resolutions, and causal attention with KV caching for reduced latency.

  • EasyControl Ghibli

    jamesliu1217 · Text-to-Image

    A Hugging Face Space for style transfer or image generation inspired by Studio Ghibli aesthetics, currently experiencing configuration errors with GPU deployment.

  • ComfyUI Portrait Master

    Stefano Flores · Text-to-Image

    A specialized node extension for ComfyUI that streamlines portrait generation workflows with optimized prompting and sampling parameters for character-focused image synthesis.

Video & Animation (3)

  • AnimeGamer

    Tencent PCG / City University of Hong Kong · Text-to-Video

    AnimeGamer enables infinite anime life simulation by using multimodal large language models to generate contextually consistent multi-turn game states as dynamic video clips with character state updates. The system employs action-aware multimodal representations decoded through video diffusion models to maintain visual coherence across player interactions with anime characters.

  • SkyReels-A2

    Skywork AI, Kunlun Inc. · Text-to-Video

    Controllable video generation framework that assembles multiple reference images (characters, objects, backgrounds) into synthesized videos using text prompts while maintaining per-element fidelity through joint image-text embedding and temporal feature consistency in diffusion transformers.

  • Segment Any Motion in Videos

    UC Berkeley / Peking University · Video Understanding

    Unsupervised moving object segmentation in videos combining long-range trajectory motion cues with DINO semantic features and SAM2 for pixel-level mask generation. Achieves state-of-the-art performance on multiple datasets including challenging scenarios with camera movement and motion blur.

3D & Spatial (1)

  • HSMR - Human Skeleton and Mesh Recovery

    UT Austin & Zhejiang University · 3D Reconstruction

    End-to-end transformer-based method for reconstructing 3D human pose and shape from single images using biomechanically accurate skeleton models, achieving competitive performance on benchmarks while outperforming SMPL-based methods on extreme poses by enforcing realistic joint angle constraints.

Developer Tools (1)

  • ADK Python

    Google · AI Infrastructure

    Python library for Google's Android Development Kit, providing SDK bindings and tools for Android application development. Enables programmatic interaction with Android development workflows and device management.

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