Quanta AI Labs

05 / Jan 27 – Feb 2, 2025 · 8 RELEASES

AI News Week 5, 2025

Week 5 2025 AI releases.

Vision & Image (1)

  • SwiftSketch

    Tel Aviv University, MIT CSAIL, Reichman University · Text-to-Image

    A diffusion-based model that generates resolution-independent vector sketches from images by denoising in stroke coordinate space, trained on ControlSketch, a new synthetic dataset of 35,000 image-sketch pairs across 100 object categories generated using an optimization-based technique.

Video & Animation (2)

  • Phantom

    ByteDance Research · Text-to-Video

    Phantom is a subject-consistent video generation model that uses cross-modal alignment to preserve identity and attributes from reference images while generating videos from text prompts. It supports single and multi-reference inputs for maintaining consistency across faces, objects, animals, and virtual characters.

  • SkyReels-V1

    SkyworkAI · Text-to-Video

    Video generation model from SkyworkAI available on GitHub. Specific capabilities and technical details require repository access to verify.

3D & Spatial (1)

  • Pippo

    Meta Reality Labs · 3D Generation

    Diffusion Transformer model that generates 1K resolution multi-view images of humans from a single photo in one forward pass, trained on 3B human images and fine-tuned on 2.5K studio captures using ControlMLP and attention biasing techniques.

Agents & Automation (2)

  • AutoAgent

    HKUDS · Agent Framework

    An agent framework for automated task execution and multi-step reasoning. Enables autonomous agents to decompose complex problems and interact with tools dynamically.

  • AI Co-Scientist

    Google Research · Agent Framework

    Google presents an AI system designed to accelerate scientific research by automating hypothesis generation, experimental design, and analysis tasks. The co-scientist framework integrates large language models with domain-specific tools to assist researchers across multiple scientific domains.

Research & Safety (2)

  • Majorana 1 Chip

    Microsoft · Survey & Overview

    Microsoft announces Majorana 1, a quantum computing chip designed to advance topological quantum error correction using Majorana zero modes. The chip represents a significant step toward practical, scalable quantum computers with improved error rates.

  • Dynamic Concepts

    Snap Research · Research Paper

    Research on dynamic concept learning that enables AI models to adapt and personalize their understanding based on user-specific or contextual variations. The work explores how concepts can be flexibly represented and updated rather than treated as static entities.

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