Google doesn’t have a great track record when it comes to creating images with artificial intelligence. Recently, a scandal erupted over the image generator built into Gemini, which incorrectly created images of people, mistaking their race. After many complaints, the tech giant removed the face generation function and sent it “for repair”. While we wait, the company has released an improved Imagen 2 imaging tool as part of its Vertex AI developer platform. However, there is a nuance – the tool is still focused on business.
Imagen 2 can create and edit images with text prompts, similar to DALL-E and Midjourney. The tool can display text, emblems and logos in multiple languages, optionally overlaying these elements onto existing images such as business cards, clothing and products.
After first launching in preview mode, image editing with Imagen 2 is now available in Vertex AI along with two new capabilities: painting and colouring. The painting and colouring functions, which have long been offered by other popular image generators, including DALL-E, can be used to remove unwanted parts of the image, add new components, and expand the boundaries of the image to create a wider field of view.
But the real upgrade to Imagen 2 is what Google calls “text to live images.” The tool can now create 4-second videos from text prompts, similar to AI-powered clip-creation tools. In keeping with the corporate focus of the model, Google offers live images as a tool for marketers and creatives, such as a GIF generator for advertising that showcases nature, food and animals – themes that Imagen 2 has been perfectly trained for.
Google claims that Live View can capture “various angles and movements” while “maintaining consistency.” But for now, they have a low resolution: 360×640, and the company promised that this will improve in the future.
To reduce concerns about the possibility of deepfakes, Google said Imagen 2 will use special SynthID technology developed by Google DeepMind to apply invisible cryptographic watermarks to live images. They are resistant to editing, including compression, filters, and colour tone adjustments, but require a separate tool to detect them.
Google also emphasizes that the generation of live images “will be checked for security purposes.” “The Imagen 2 model in Vertex AI did not experience the same problems as the Gemini application. We continue to actively test and interact with our customers,” company representatives noted.
Read also:
- Google Pixel’s market share in North America is almost 5%
- Robots of the future: will artificial intelligence get a body?