Quanta AI Labs

36 / Sep 1 – 7, 2025 · 10 RELEASES

AI News Week 36, 2025

Week 36 2025 AI releases.

Language Models (1)

  • Ming-Lite-Omni

    inclusionAI · Multilingual

    Unable to analyze - page content unavailable. Ming-Lite-Omni appears to be a model hosted on Hugging Face, likely a multimodal or language model variant based on naming convention.

Video & Animation (5)

  • HunyuanVideo-Avatar

    Tencent Hunyuan / Tencent Music Entertainment / Lyra Lab · Text-to-Video

    Multimodal diffusion transformer model for generating high-fidelity audio-driven human animation with character consistency, emotion control, and multi-character dialogue support. Introduces character image injection, Audio Emotion Module, and Face-Aware Audio Adapter for precise facial animation and independent character control.

  • SeedVR2

    ByteDance Seed, Nanyang Technological University · Video Editing

    One-step diffusion-based video restoration model using adversarial post-training with adaptive window attention for efficient 1080p restoration. Achieves comparable or better performance than multi-step approaches while maintaining temporal consistency and computational efficiency.

  • Any-to-Bokeh

    Vivo Mobile Communication Co., Ltd & Zhejiang University · Video Editing

    One-step video bokeh framework using multi-plane image (MPI) guided diffusion to generate temporally coherent, depth-aware bokeh effects with controllable focal planes and blur intensity. Leverages pre-trained Stable Video Diffusion with progressive training strategy for consistent results across diverse scenes.

  • OmniSync

    Kuaishou Technology (Kling Team), Renmin University of China, Tsinghua University · Video Editing

    Universal lip synchronization framework using Diffusion Transformers with mask-free training, progressive noise initialization, and Dynamic Spatiotemporal Classifier-Free Guidance to maintain identity consistency while precisely editing mouth regions. Introduces AIGC-LipSync Benchmark for evaluating lip sync in AI-generated videos across diverse visual scenarios including stylized characters and occluded faces.

  • LayerFlow

    HKU, DAMO Academy (Alibaba), Hupan Lab · Text-to-Video

    LayerFlow is a unified diffusion transformer model for layer-aware video generation that decomposes videos into transparent foreground, clean background, and blended scenes using per-layer prompts. It supports multi-layer generation, decomposition, and conditional layer synthesis through a multi-stage training strategy combining low-quality video data with high-quality layered image annotations.

Developer Tools (4)

  • OneFileLLM

    jimmc414 · Inference Engine

    A lightweight implementation for running LLMs with minimal dependencies, designed to fit a complete LLM system into a single file for educational and practical purposes.

  • ComfyUI-8iPlayer

    Kartel-ai · AI Infrastructure

    A ComfyUI plugin that integrates 8i volumetric video player capabilities, enabling real-time playback and manipulation of volumetric content within the ComfyUI node-based workflow system.

  • MCP Server Chart

    AntVis · AI Infrastructure

    A Model Context Protocol (MCP) server implementation that enables chart and visualization generation capabilities. Allows LLMs to create and manipulate data visualizations through standardized protocol interfaces.

  • ComfyUI-FLOAT

    yuvraj108c · AI Infrastructure

    A ComfyUI extension for floating-point precision control and optimization in image generation workflows. Enables advanced numerical precision handling for diffusion model inference.

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