Google announces image generation AI 'Imagen 4', capable of generating images up to 2K and a high-speed version 10 times faster than Imagen 3 is also scheduled to be released



Google announced its image generation AI ' Imagen 4 ' on May 21, 2025, Japan time.

Imagen - Google DeepMind

https://deepmind.google/models/imagen/

Imagen 4 Model Card - Imagen-4-Model-Card.pdf
(PDF file) https://storage.googleapis.com/deepmind-media/Model-Cards/Imagen-4-Model-Card.pdf

Fuel your creativity with new generative media models and tools
https://blog.google/technology/ai/generative-media-models-io-2025/

Imagen 4, Veo 3: Google's latest media generation models
https://blog.google/intl/ja-jp/company-news/technology/aigenerative-media-models-io-2025/

Imagen 4 can generate images up to 2K in resolution, allowing you to generate high-quality images ranging from photorealistic to illustrative.




Here is an example of a photo-like image generated by Imagen 4. Both the person and the background are reproduced quite realistically.



Image of a bird taking flight.



This image looks like a butterfly wing photographed with a macro lens, showing even the fine details of the wing's pattern.



It is also possible to generate illustration-like images. The example below is a five-panel comic strip generated by specifying the situation and dialogue for each panel.



Below are the results of ' GenAI-Bench ', which evaluates the performance of generative AI. Imagen 4 recorded a higher score than Stable Diffusion 3.5 Large and FLUX.1.



Below is a graph of the GenAI-Bench score (vertical axis) and generation speed (horizontal axis). Imagen 4 was able to generate high-quality images in a short time.



Imagen 4 is already available in the Gemini app, Whisk, Vertex AI, and Google Workspace, and a high-speed version that is up to 10 times faster than Imagen 3 is coming soon.



Related Posts:

in Software, Posted by log1o_hf