Imagen

A family of text-to-imagetext-to-image model, n.: An AI system that creates a picture from a written description, such as 'a fox reading a newspaper, in watercolor', usually within seconds. diffusion modelsdiffusion model, n.: An AI that learns to create images, sound or video by practicing noise removal: shown pictures buried under static, it learns to clean them up, then builds new ones starting from pure static. from Google. The original Imagen (May 2022) used a frozen T5-XXL language model as its text encoder and a cascade of pixel-space diffusion models: a base model generating 64 × 64 images followed by two text-conditioned super-resolutionsuper-resolution, n.: Making a small or blurry image larger and sharper by having AI fill in plausible fine detail, rather than just stretching the pixels. models upscaling to 256 × 256 and 1024 × 1024. Its paper reported a zero-shot COCO FIDFréchet inception distance, n.: A score for image generators that compares the overall statistics of their pictures with those of real photos, as seen by an image-recognition network; lower is better. of 7.27, introduced the DrawBench prompt benchmark and argued that the text encoder matters most: enlarging the language model raised image quality and prompt alignment far more than enlarging the diffusion model. Google initially withheld the model, citing risks of misuse and social bias, then shipped successors in its products and cloud platform: Imagen 2 (December 2023), Imagen 3 (2024) and Imagen 4 (May 2025). Gemini-based image models (Nano BananaNano Banana, prop. n.: The nickname, later Google's official branding, for its Gemini image models, a viral hit in 2025 for editing photos from simple instructions while keeping people looking like themselves.) then took over much of Google's image generation, and Google retired the Imagen models from its Gemini API on 17 August 2026, directing developers to the Gemini image models.