Google has released its latest text-to-image generator, Imagen 3, to users in the US, marking a significant update in the company’s AI offerings. The model is available in the company's AI Test Kitchen and its Vertex AI platform.
According to Google, Imagen 3 represents the company’s highest-quality model to date, boasting improvements in detail, lighting, and composition. The model is designed to generate images with fewer artifacts and a greater ability to capture small nuances. Users can create various styles, from photorealistic images to abstract artworks, with much less effort in prompt engineering thanks to the model’s improved natural language understanding.
Google’s decision to release Imagen 3 without much fanfare may be influenced by the industry's excitement surrounding Black Forest Lab’s recently launched Flux-1. This open-source model has captivated the AI community with its impressive capabilities, especially when paired with realism LoRA. While Google’s Imagen 3 is a notable advancement, the landscape has shifted drastically since it was first announced at Google I/O in May.
Still, Imagen 3 is big upgrade from its predecessor. It has enhanced capabilities that include better text rendering, making it possible to use the tool for creative applications such as stylized cards and presentations. Google says it plans to bring over features from Imagen 2, like inpainting and outpainting, for more advanced image editing .
The company emphasized its focus on safety and ethics in the development of Imagen 3. This includes extensive filtering and data labeling processes aimed at minimizing harmful content in the training dataset. Additionally, Google has integrated its SynthID watermarking technology into Imagen 3, embedding imperceptible markers in generated images to facilitate their identification as AI-created content.
This approach stands in stark contrast to models like Flux, which offer users the flexibility to fine-tune the AI with custom datasets and do not include built-in watermarking. While this openness has fueled excitement in some quarters of the AI community, it also raises questions about potential misuse and the challenges of identifying AI-generated content.