Google introduced a new text-to-image model Imagen 4, which is now available for preview in the Gemini API. From June 24, users can test it for free in Google AI Studio. The model allows creating up to four images sized 1024 by 1024 pixels per request and accepts prompts up to 480 tokens.
The cost of one image is about $0.04, and the higher quality version Imagen 4 Ultra costs approximately $0.06 per image. Both versions use SynthID watermarks and provide almost instant image creation. Imagen 4 Ultra ensures more accurate prompt execution and allows exporting images in 2K format via Vertex AI.
Developers can use the same endpoint “/generate” as for the Gemini models and can request limit increases during the test period. The standard limit is 20 requests per minute for one project.
Early users are already sharing results on the social network X, noting clearer text and fewer artifacts compared to Imagen 3. The launch of Imagen 4 expands the Google DeepMind toolset, which already includes Imagen, Veo, and the Gemini model family.
