Google develops a generative AI of images from text with custom artistic styles

June 21 (Portaltic/EP) –

Google is developing an Artificial Intelligence (AI) that generates images from text called StyleDrop, which allows them to be created with personalized artistic styles, as well as adding fonts that incorporate characteristics similar to those of these images.

The Senior Vice President of Research, Technology and Society at Google, James Manyika, and the Vice President of Google Creative LabRobert Wong, have presented some of the projects they are working on from Google’s Creative Lab.

They have done so within the framework of the Cannes Lions International Festival of Creativity (France), where they have focused on everything that AI can offer and the focus that the company is currently giving it with solutions such as StyleDrop.

StyleDrop is a project that Google Research is currently working on and that allows you to generate images from text in any style, capturing “nuances and details of a style proportioned” and personalized by the user, according to the company.

Google has insisted that this tool can deliver “impressive results even when the user provides a single image specifying the desired style.” By example, ‘3D rendering style’, so that it generates logos and icons that are creative and stylistically consistent with the Android brand image in Google. To do this, this tool, which is based on the AI ​​Muse model, mimics color and shading schemes, design patterns.

For its part, the company has pointed out that this tool “convincingly outperforms” others for generating images from text, such as Textual Inversion -based on Image-, stable diffusion either Dream Booth.

Google has also mentioned the latter in Cannes Lions. Specifically, it is presented as a solution that identifies the subject and generates photorealistic images by placing it in other contexts, with other poses, with accessories and from other perspectives.

From three sample images, DreamBooth can also show that subject in different art styles. For example, a dog painted as if he had done it Van Gogh -with bright colors and sharp brushstrokes- or rembrandt -with sidelights and shadows-.

Likewise, this generative AI tool can take a specific image (for example, a dog) and create other images in which it formulates hybrid beings while maintaining the characteristics of the initial animal. Thus, you can offer the image of a bear, a lion or a hippopotamus with the characteristic features of a Chow chow.

Finally, the managers have referred to a pilot and experimental tool called AI-Replywhich uses Palm 2 and MakerSuite and that it can generate draft responses to user comments on its YouTube platform.

In this sense, they have commented that the platform works with a manager who evaluates, selects and edits the best possible response options to offer them with a friendly tone and maintaining the brand image of Google.

The company has qualified, finally, that it advocates “build AI responsibly” and believes that “it must be a collective effort involving researchers, social scientists, industry experts, governments and ordinary people.”