Visual AI Digest: Model Rivals, Unified Editing & Complex Action Control

Research

New skills breed control for text-to-video generation of complex actions

This paper tackles a real blocker in video models: animating complex, long-range actions like opening a fridge or folding laundry, which current models drop the ball on.

UniEdit: A unified model for image and video editing

This unifies single-image and video editing into one model, eliminating the need for separate pipelines. It's a significant step towards more coherent and scalable editing workflows.

Towards complex 3D scene generation from text

Introduces a new framework for generating complex 3D scenes from text, pushing beyond flat images and simple videos towards richer, more interactive visual content.

Text-driven pose generation for diverse human characters

This work focuses on generating high-quality, diverse human poses from text, a key step for realistic character animation in videos without relying on reference videos.

News

Alibaba & ByteDance launch new AI models to compete with OpenAI's Sora

Alibaba and ByteDance are aggressively entering the AI video race, directly challenging OpenAI's Sora with new models and pricing. This signals a major escalation in the global competition for video AI dominance.