Google Brings Personalized Image Generation to Free Gemini Users

AWS and Vexcel, an aerial imagery provider operating across 45+ countries, developed a multimodal AI system that converts billions of aerial images into natural-language-searchable data without requiring per-feature model training. The system uses embedding models, LLM captioning, and vector search to index imagery once and query it with plain English. Amazon Nova Multimodal Embeddings delivered the highest F1 scores in their evaluation, and the work evolved into Vexcel Intelligence, a commercial searchable imagery product.

Google DeepMind's Gemma 4 model family is now available on Amazon Bedrock, offering three instruction-tuned variants ranging from 2.3B to 30.7B parameters. The models support reasoning, function calling, and multimodal input while running on AWS infrastructure with data protection guarantees. Organizations can access open-weight models through a managed service without hosting infrastructure themselves.

Researchers from UC Berkeley, Princeton, EPFL, and Databricks introduced PixelRAG, a retrieval system that bypasses traditional text parsing by rendering web pages as screenshots and indexing them directly for vision-language models. Tested on 30 million Wikipedia screenshot tiles, PixelRAG improved accuracy by up to 18.1% over text-based RAG systems and reduced token costs by 10x. The approach addresses fundamental information loss in conventional HTML-to-text conversion pipelines.

Google DeepMind introduced Gemma 4 12B, a multimodal AI model designed to run on consumer laptops with 16GB of RAM. The model uses an encoder-free architecture that processes vision and audio inputs directly into the language model backbone, reducing latency and memory overhead. Performance approaches the larger 26B model while maintaining a smaller footprint, and it is released under an Apache 2.0 license.