Google introduces Imagen 4 — a text-to-image model with improved text rendering

Google has announced a new image generation model, Imagen 4, along with its advanced version, Imagen 4 Ultra. According to the company, the model significantly improves text rendering quality — a longstanding challenge for similar tools.

Trained on extensive datasets, Imagen 4 not only generates detailed visuals from text prompts but also accurately reproduces written text, whether it’s standalone phrases or lettering integrated into objects.

For example, if a prompt says that a spaceship should have “Stardust” written on it, the model actually renders a readable inscription on the hull — rather than a distorted jumble of letters, as was often the case with previous models.

Google released two versions:

• Imagen 4 – a general-purpose model for a wide range of users and standard prompts.
• Imagen 4 Ultra – a more precise and powerful version designed for generating visuals that closely align with the input prompt. It requires slightly more resources and is priced higher.

The model is available:

• For free via Google AI Studio.
• Through the Gemini API – at $0.04 per standard image and $0.06 for Ultra-generated images.
• Not yet in the Gemini chatbot, though Google says integration is coming soon. AI Studio subscriptions for generation will roll out in the coming weeks.

The previous version, Imagen 3, was released in May 2024 and already made notable progress in quality and text accuracy. Imagen 4 builds on that foundation — with a strong focus on realism, detail, and clear, legible text in generated images.