Google's Imagen 4 text-to-image model promises 'significantly improved' boring images

Google has introduced its latest text-to-image model, Imagen 4, which promises significantly improved text rendering over its predecessor, Imagen 3. The company has also launched a deluxe version called Imagen 4 Ultra, designed to follow more precise text prompts at an additional cost. Both models are available for paid preview in the Gemini API and limited free testing in Google AI Studio. The main Imagen 4 model is priced at $0.04 per image and is described as suitable for most tasks. Imagen 4 Ultra, on the other hand, is priced at $0.06 per image and is intended for tasks that require precise instruction following. Google has showcased a range of images generated by Imagen 4 Ultra, including a three-panel comic and a vintage travel postcard, which demonstrate the model's ability to follow text prompts accurately. However, the images generated by Imagen 4 lack charm and appear highly machine-generated, despite being of good quality. The model's performance is considered a mild improvement over its predecessor, but it fails to impress, particularly when compared to market leaders like Dall-E 3 and Midjourney 7. The public's enthusiasm for AI art seems to be waning, with the main use case being spammy ads on social media or at the bottom of articles. Overall, Imagen 4 and Imagen 4 Ultra demonstrate Google's continued efforts to improve its text-to-image models, but the results are not yet groundbreaking.