Developed in collaboration with Google DeepMind, Imagen 2 promises higher image quality than the first version as well as new features such as the ability to generate text and logos on images. Thomas Kurian, the head of Google Cloud, emphasized during a press conference that this new AI can be used to create images with overlaid text, for example in advertising.
An AI adapted for advertising
An advance that puts Imagen 2 in direct competition with other image generation models such as DALL-E 3 from OpenAI and Titan Image Generator from Amazon. An important feature of Imagen 2 is the ability to generate text in multiple languages, including Chinese, Hindi, Japanese, Korean, Portuguese, English and Spanish. In addition, the model can embed logos on various products and surfaces, which represents a significant advance in the field of image creation. Advertisers will definitely appreciate it!
For traceability, the images generated by Imagen 2 contain the SynthID watermark developed by DeepMind. The fact is that this launch is surrounded by controversy. Google has not disclosed the data used to train this model, which obviously will not reassure anyone about intellectual property rights and creator protection. This opacity contrasts with the strategy of the first version of Imagen, where Google disclosed the use of a version of the public LAION dataset.
© Google
The legal question of whether AI providers can use public data, including proprietary data, to train and then commercialize a model remains unresolved. There are several lawsuits underway in which the publishers of these services are defending themselves by citing fair use. In the meantime, Google appears to be erring on the side of caution and maintaining discretion over the data it uses.
Additionally, unlike some companies like Stability AI and more recently OpenAI, Google does not offer creators the ability to remove their content from training databases or a compensation mechanism. This lack of recognition of creators' rights is a cause for concern in the industry.
Given the undeniable technological progress, questions arise about the use of training data and respect for the rights of authors. However, Google is participating in current discussions at the highest levels, be it in the US or Europe, to find common ground on these very sensitive issues.
🟣 To not miss any news in Journal du Geek, subscribe to Google News.