# AI Veo and Project Astra: What New Features Were Unveiled at Google I/O

Sharing the most interesting new developments in the AI field that were announced this year at Google's annual I/O conference.

**[AI Assistant Project Astra](<https://www.theverge.com/2024/5/14/24156296/google-ai-gemini-astra-assistant-live-io>)**  
This is an AI assistant that operates in real time. It can recognize objects, remember where things are located, answer questions, and help solve various tasks. Project Astra is expected to be available not only on smartphones but also on smart glasses with cameras.

**[Gemini integrates into Gmail](<https://www.theverge.com/2024/5/14/24156500/google-gmail-ai-gemini-summaries-contextual-smart-reply>)**

AI Gemini will be able to answer user questions, suggest ready-made responses, and perform routine tasks in the background. Additionally, the artificial intelligence will be able to automate the processing of numerous email attachments and organize them in Google Drive cloud storage.

**[AI Model Gemini 1.5 Flash](<https://www.theverge.com/2024/5/14/24155511/google-gemini-ai-flash-1-5-pro-updates-io>)**[](<https://www.theverge.com/2024/5/14/24155511/google-gemini-ai-flash-1-5-pro-updates-io>)Gemini 1.5 Flash will work for "narrow, high-frequency tasks with low latency." Gemini 1.5 Flash may be the best option for real-time customer responses or quick image generation, while Gemini 1.5 Pro can read and summarize research papers.

**[AI Veo for creating realistic videos](<https://www.theverge.com/2024/5/14/24156255/google-veo-ai-generated-video-model-openai-sora-io>)**  
Veo is the latest generative video model with artificial intelligence — it can generate "high-quality" videos with a resolution of 1080p lasting more than a minute in a wide variety of styles. Veo has "advanced understanding of natural language," which allows the model to understand cinematic terms.

**[AI Image Generator Imagen 3](<https://techcrunch.com/2024/05/14/google-launches-new-video-and-image-generation-tools/>)**

Imagen 3 demonstrates a more accurate understanding of test prompts, leading to the creation of more creative and detailed images compared to the Imagen 2 model.
