Google launches Imagen 4 with enhanced image generation features

Summary

• Google releases Imagen 4 for free testing via AI Studio for a limited time
• New model boasts faster performance, photorealistic detail, and better text rendering
• Premium Ultra version aims to rival top-tier competitors like DALL-E 3 and Midjourney

Google has unveiled Imagen 4, its most advanced text-to-image generation model to date, offering free access for a limited time through Google AI Studio. The model is also available via the Gemini API, with paid access starting at $0.04 per image and rising to $0.06 for the high-precision Imagen 4 Ultra variant.

Imagen 4 marks a major leap from its predecessor, delivering photorealistic images with fine-grained details such as water droplets, animal fur, and varied textures. It supports multiple styles—ranging from abstract and impressionist to animated—as well as text rendering capabilities suitable for posters, comics, and greeting cards. Images can be generated in multiple aspect ratios and up to 2K resolution.

The Ultra version, designed for users requiring closely matched visual outputs from complex prompts, is pitched as competitive with top tools like OpenAI’s DALL-E 3 and Midjourney 7. However, early feedback is mixed: while some praise its speed and detail, others criticise the results as generic and overly mechanical.

Performance-wise, Imagen 4 is said to be up to 10 times faster than Imagen 3 in certain modes, with safety measures such as SynthID—a digital watermark that allows for the tracing of AI-generated content—embedded in every image. Nonetheless, the model reportedly struggles with more intricate compositions, such as centring subjects or accurately depicting small faces and thin elements.

Google plans to roll out Imagen 4 across its suite of products, including Gemini, Slides, Docs, and Whisk, with some third-party platforms like Freepik and Pollo AI already offering early access.

While ethical safeguards have been included, concerns remain around potential misuse of deepfakes and copyright infringement, as well as persistent bias in training data derived from web sources. Some users have also voiced fatigue over AI-generated art, especially where it fuels online spam.