Realtime AI heads, new AI image tools, long AI videos, Veo2 is free, boxing robots, AI horses

# Weekly AI Tech Roundup: Cutting-Edge Innovations and Tools

This week’s AI developments showcase remarkable advancements across image generation, video creation, 3D modeling, and humanoid robotics. From bite-sized tools to comprehensive enterprise solutions, the AI landscape continues to evolve at breakneck speed.

## Image and Video Generation Breakthroughs

### Uno by ByteDance: Multi-Reference Image Generation
ByteDance’s new Uno tool allows users to create images with multiple reference objects or characters. Unlike similar tools, Uno excels at maintaining accuracy and consistency when combining different elements. Use cases include:
– Adding logos to clothing
– Combining multiple objects in one scene
– Generating characters in different artistic styles
– Creating virtual fashion models
– Face swapping and character cloning

ByteDance has released a Hugging Face demo and GitHub repo, requiring minimum 16GB of memory to run locally.

### One-Minute Video Generation
A new AI can create coherent one-minute videos from detailed text storyboards that specify each scene. While not production-ready (with issues in text legibility and edge artifacts), this prototype demonstrates how future AI could generate entire episodes with multiple coherent scenes.

The system builds upon the Cog Video base model (normally limited to 3-second clips) with added “test time training layers” that maintain style and character consistency throughout longer sequences.

### V2 Now Free in Google AI Studio
Google’s V2 video generation model is now freely available in Google AI Studio. Users can generate 8-second videos by selecting parameters like aspect ratio and optional negative prompting. The system also offers auto-save functionality to Google Drive.

## 3D Modeling and Vector Graphics

### Hollow Part: Complete 3D Object Segmentation
This tool breaks down 3D objects into smaller complete parts, even filling in hidden elements. The two-stage approach first identifies visible parts, then uses a special 3D shape diffusion model to complete any missing areas. Applications include improved downstream editing like changing textures or sizes of specific components.

A free Hugging Face demo and GitHub repo are available.

### OmniSVG: High-Quality Vector Graphics
OmniSVG creates high-quality SVG (scalable vector graphics) from