Google has released Imagen 4, a new text-visual image generation model that the company says delivers "significantly improved text rendering" over the previous version. It also introduced a variant, Imagen 4 Ultra, with increased accuracy in matching query instructions.
The company first announced the new model at Google I/O in May of this year. Now, both models are becoming available for limited testing in Google AI Studio, as well as through a paid subscription to the Gemini API.
The regular version of Imagen 4 costs $0.04 per image and is designed for most popular generation tasks. The Ultra version is aimed at more demanding scenarios with exact adherence to the description and costs $0.06 per image.
In the examples provided by Google, Imagen 4 Ultra was able to generate a comic book page with a complex prompt and a retro postcard with a Kyoto landscape. However, the realistic images still have a characteristic "artificial" look that makes it easy to recognize that they were created by a neural network.
Google positions Imagen 4 as a direct competitor to other image generators, including Dall-E 3 and Midjourney 7.