Imagen is Google's earlier family of text-to-image models, first described in 2022 and now shut down in the Gemini API in favour of Nano Banana.
In plain words
Imagen paired a large language model, which understands the prompt, with diffusion models, which draw the picture step by step. Its researchers found that making the text model bigger improved results more than making the image model bigger. Later versions followed, including Imagen 3 in 2024 and Imagen 4, which reached the API in 2025. Google now calls Imagen a legacy model and tells developers to switch to Nano Banana.
A quick example
Imagine a prompt describing a corgi in a sunflower field at sunset. Imagen's text encoder turns the words into numbers, and the diffusion model gradually turns noise into a matching photo-like image.
Where you'll meet it
- Text-to-image research and the Gemini API, before Nano Banana.
Related
Sources
- PaperPhotorealistic Text-to-Image Diffusion Models with Deep Language Understanding, Saharia et al. (arXiv) · read 29 Sept 2026
- DocsGenerate images using Imagen, Gemini API docs · read 2 Nov 2026
- DocsVeo and Imagen 3, Google · read 29 Sept 2026
- DocsGemini deprecations, Gemini API docs · read 2 Nov 2026
Checked by WhatIsAI against the sources above · 2 Nov 2026. Spotted an error? Report it.