Google Cloud aims to fulfill its previous promise of powerful image AI with the arrival of Veo and Imagen 3 to the Vertex AI platform. These models promise to significantly accelerate enterprise graphics asset production and keep costs down.
Developed by Google’s AI factory DeepMind, Veo is now available in ‘private preview’ on Vertex AI, the machine learning platform available exclusively on Google Cloud. This means that for now, it is only for a select group of customers, although Google will further roll out the service over time.
The model manages to produce videos in high definition based on textual prompts or images as sources. Different styles are possible, such as realistic videos or animations. For example, companies can animate existing assets, or create completely new videos from a single textual prompt and further customize them if necessary until the result is to their liking.
Imagen 3 is a text-to-image generator and is generally available to all Vertex AI users starting next week. The tool makes it possible to use simple text as input to create photorealistic images whose details have a high degree of fidelity. Google emphasizes this last part because image generators often struggle with this, sometimes conjuring up too many fingers, generating nonsensical inscriptions, or creating bizarre-looking utensils.
True-to-life images
Imagen 3 also promises true-to-life lighting and fewer artifacts compared to previous versions. By artifacts, the company means objects in the image that are incorrect or out of place, as well as incorrect use of colors, lights, or lack of sharpness. Adding your own brand or corporate identity is also possible, as is applying different styles such as a cartoon or painting look.
Tip: Better, faster, more affordable: Google Cloud significantly expands Vertex AI portfolio
What should make this tool especially interesting for marketers, product designers, creatives or PR personnel is the ability to adjust an image at the detail level. So it goes far beyond generating a few images and hoping there is something useful among them: an ‘okay’ result can be tweaked until it is really good. Also useful: providing a text prompt with an attached image for reference. For example, a picture of a dress and asking the AI to generate an image of a woman wearing it.
Provided with a digital watermark
Google understands that forgeries and deepfakes pose a risk (intentionally or unintentionally) with such powerful software. Both Veo and Imagen 3 are therefore equipped with a digital watermark via DeepMind’s SynthID that should counter disinformation, prevent harmful content and protect user information. The latter, by the way, is no unnecessary luxury.
Google also offers copyright indemnity, which means that if a party (wrongly) claims that their copyright has been infringed upon using these tools, Google will take care of all legal consequences. Provided, of course, that the tools have not been misused.
Also read: Google shows off new set of AI tools for creating advanced videos and images