Official Google DeepMind announcements or release notes

Google DeepMind's Imagen 3 Outperforms DALL-E 3, Now Available via ImageFX to U.S. Users

Aug 15, 2024, 05:42 PM

Google DeepMind has unveiled Imagen 3, an advanced AI image generation model that excels in creating high-quality, photorealistic images from complex text prompts. The model, which uses Tensor Processing Units (TPUs) for efficient training, is trained on a large curated dataset and incorporates safety filters, image deduplication, and synthetic captioning using Gemini models. Imagen 3 has outperformed other state-of-the-art models such as DALL-E 3, Midjourney v6, and Stable Diffusion 3 in human evaluations. The AI model is now available to all U.S. users via ImageFX, sparking debates on AI ethics and creativity. Despite its capabilities, Imagen 3 faces challenges with numerical reasoning and action depiction, and some users criticize its strict content filters.