Image Generation

Today's Stories

1. 1

   Google releases DiffusionGemma AI model that generates text 4 times faster

   Google DeepMind launched DiffusionGemma, an experimental AI model that generates text blocks of 256 words simultaneously instead of typing one word at a time like ChatGPT. The model uses the same technique that image generators like Stable Diffusion use - starting with random noise and refining the entire output together. It can produce over 1,000 tokens per second on high-end hardware and 700+ on consumer graphics cards.

   Future AI writing tools and chatbots could respond much faster, especially when running locally on your computer without internet connection.

2. 2

   Microsoft proves detailed image captions beat massive datasets for AI training

   Microsoft Research created Lens, a text-to-image AI with only 3.8 billion parameters that matches much larger competitors by using 800 million detailed image descriptions generated by GPT-4 instead of simple web captions. The approach cuts training costs significantly while maintaining quality. The code and model weights are available as open source.

   Smaller companies can now build competitive image-generating AI tools without Google or OpenAI's massive computing budgets.

3. 3

   Apple announces AI features for iPhones and iPads at WWDC

   Apple revealed new AI capabilities for iOS and iPadOS at its developer conference, including chatbot features, text summarization, and image generation tools. Most features mirror existing capabilities already available on Android phones and in ChatGPT apps. The company focused on integrating AI into its Shortcuts app and Safari browser.

   iPhone users will get AI features similar to what Android users already have, but tightly integrated into Apple's ecosystem of apps.