AI Timeline #7: Stable Diffusion SDXL 0.9, Zeroscope v2 Text-to-Video & More

Table of Contents

1. Introduction
2. ZeroScope V2 TechSoup Video Model
   • 2.1 ZeroScope V2 XL Model
   • 2.2 ZeroScope V2 576w Model
   • 2.3 ZeroScope Dark V2 Model
3. Stable Diffusions
   • 3.1 Beta Release of Stable Diffusions' XL 0.9 Base Model
   • 3.2 Comparisons with Older Versions
4. Large Language Models (LLM) News
   • 4.1 Textbooks as Training Data
   • 4.2 Cosmos 2: Towards Artificial General Intelligence
   • 4.3 MBT30B Chat: A Commercially Usable LLM
5. Motion GPT: Generating High-Quality Motions
6. Browsing Mode on ChatGPT
7. Unity Muse: AI Tool for Video Game Creation
8. Arrow Palm: Speech-to-Speech Translation
9. Limitations of Transformers in Complex Tasks
10. Drag Diffusion: Replicating Dragon's Image Editing
11. Dream Editor: Text-Driven 3D Scene Editing
12. Chatbots as Teachers at Harvard
13. AI-Based Upper Body Tracking for Vtubers
14. Stability AI's Resignations
15. Punchable AR Experience for Calorie Burning
16. Doraemon and the Prediction of AI Image Generation

ZeroScope V2 TechSoup Video Model

ZeroScope V2 TechSoup Video Model is one of the most exciting developments in the field of AI this week. This collection includes three models: the ZeroScope V2 XL model, the ZeroScope V2 576w model, and the ZeroScope Dark V2 model. These models are trained to perform different tasks and offer different resolutions and frame rates.

ZeroScope V2 XL Model

The ZeroScope V2 XL model allows users to generate videos at a resolution of 1024x576, with a frame rate of up to 24 FPS. It offers a high-resolution output, making it suitable for detailed video generation.

ZeroScope V2 576w Model

The ZeroScope V2 576w model is a lighter version compared to the XL model. It is trained to generate videos at a resolution of 576x320, with a frame rate of up to 24 FPS. While it offers a lower resolution, it still provides impressive video generation capabilities.

ZeroScope Dark V2 Model

The ZeroScope Dark V2 model focuses on generating videos at a frame rate of 30 FPS, although the resolution is slightly lower at 448x256. It is specifically designed to handle low-light conditions, making it ideal for generating videos in dark environments.

All three models in the ZeroScope V2 TechSoup Video Model collection were trained on a large dataset of 9923 clips, with 29,769 tagged frames. This extensive training dataset ensures that the models can produce high-quality video outputs.

To use the ZeroScope V2 TechSoup Video Model, users can utilize the Automatic111 text-to-video extension. More workflows and instructions can be found on the official Hooking Face model cards.

Pros of ZeroScope V2 TechSoup Video Model

• High-resolution video generation (XL model)
• Lighter model option available (576w model)
• Designed for low-light conditions (Dark V2 model)
• Extensively trained on a large dataset

Cons of ZeroScope V2 TechSoup Video Model

• Lower resolution option (576w model)
• Slightly lower resolution in the Dark V2 model

The ZeroScope V2 TechSoup Video Model collection promises exciting possibilities for video generation, whether it's for professional use or creative projects.

Stable Diffusions

Stable Diffusions has introduced its XL 0.9 base model, focusing on generating images with greater details and composition. Although still in beta, this model is available on Clip Drop, a new free service specifically created for generating images.

Beta Release of Stable Diffusions' XL 0.9 Base Model

The XL 0.9 base model is a significant step forward in image generation. It offers improved aesthetics, enhanced variation drops with controllable strength, and a new feature called "Zoom Out in Painting." With four options available, users can choose the zoom level that best suits their needs.

Comparisons between the older SD XL version and the SD XL 0.9 model highlight the significant improvements in composition, colors, and details. Users have also compared the SD XL 0.9 model with SD 1.5, showing its ability to generate high-quality images with superior details.

It is important to note that some users have attempted to generate images with SD 1.5 to match the output of SD XL 0.9 by using control net inpainting and upscaling techniques. However, the SD XL 0.9 model consistently outperformed SD 1.5 in terms of details and overall image quality.

Pros of Stable Diffusion's XL 0.9 Base Model

• Improved aesthetics
• Greater variation control
• Superior composition, colors, and details

Cons of Stable Diffusion's XL 0.9 Base Model

• Still in beta
• Long waiting times between edits due to slower processing

The XL 0.9 base model from Stable Diffusions showcases the potential for AI-generated images with enhanced quality, paving the way for exciting possibilities in digital art and visual content creation.