Quanta AI Labs

04 / Jan 20 – 26, 2025 · 11 RELEASES

AI News Week 4, 2025

Week 4 2025 AI releases.

Video & Animation (5)

  • Enhance-A-Video

    National University of Singapore, Shanghai AI Lab, University of Texas at Austin · Video Editing

    A training-free video enhancement method that adjusts temporal attention temperature in diffusion transformer models to improve cross-frame consistency and visual quality. The technique applies an Enhance Block to existing video generation models without requiring additional training.

  • CineMaster

    Dalian University of Technology, Chinese University of Hong Kong, Kuaishou Technology · Text-to-Video

    A 3D-aware text-to-video generation framework enabling precise control over object placement and camera movements in 3D space through an interactive workflow and diffusion model guidance. Introduces an automated data annotation pipeline extracting 3D bounding boxes and camera trajectories from large-scale video data.

  • On-device Sora

    EAI Lab · Text-to-Video

    An on-device implementation of video generation technology enabling efficient video synthesis without cloud dependencies. Focuses on making Sora-like capabilities accessible for edge deployment.

  • FlashVideo

    Foundation Vision · Text-to-Video

    Video generation model optimized for fast inference and efficient processing. Enables rapid video synthesis from prompts or images.

  • Step-Video-T2V

    Stepfun AI · Text-to-Video

    A text-to-video generation model that converts textual descriptions into video content. The model appears to focus on efficient video synthesis with step-by-step generation capabilities.

Audio & Speech (1)

  • InspireMusic

    Tongyi Lab, Alibaba Group · Music Generation

    A unified framework for text-guided high-fidelity long-form music generation combining an ultra-low-bitrate audio tokenizer, autoregressive LLM (based on Qwen2.5), and super-resolution flow-matching model to generate 48kHz audio with distribution-aware dual-temperature truncation sampling.

Developer Tools (2)

  • ComfyUI Webcam Node

    uetuluk · AI Infrastructure

    A ComfyUI node extension that enables real-time webcam input integration into ComfyUI workflows, allowing direct video stream processing within the node-based interface.

  • Unity Gaussian Splatting

    Aras Pranckevičius · Edge Deployment

    Integration of Gaussian splatting rendering technology into Unity game engine, enabling efficient 3D scene reconstruction and real-time rendering from point cloud data.

Research & Safety (3)

  • Artificial Cambrian Intelligence - Vision Evolution Simulator

    MIT Media Lab et al. · Research Paper

    Computational framework that evolves artificial embodied agents in physics-simulated environments to study how visual intelligence emerges from co-evolution of bodies, sensors, and behavior. Demonstrates that optimal eye designs (lenses, acuity) and brain-vision scaling laws emerge naturally under survival pressures, without manual engineering.

  • Gaussian Splatting

    Inria (GRAPHDECO) · Research Paper

    Novel 3D rendering technique using Gaussian primitives for real-time novel view synthesis, achieving state-of-the-art quality with faster training than neural radiance fields.

  • Splat

    m-niemeyer · Research Paper

    3D Gaussian splatting-based neural rendering method for novel view synthesis and real-time 3D scene reconstruction from images.

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