Google's 'StyleDrop' can generate images of your favorite style from character strings



Google has developed 'StyleDrop' that can accurately capture the contents of the style presented by the user, such as color scheme, shading, and design pattern, and output the image required by that style. It is possible to generate an image based on the input character string with

the image generation AI 'Stable Diffusion' etc., but it was difficult to output based on a specific style. 'StyleDrop' overcomes this problem.

StyleDrop: Text-to-Image Generation in Any Style
https://styledrop.github.io/



[2306.00983] StyleDrop: Text-to-Image Generation in Any Style

https://doi.org/10.48550/arXiv.2306.00983

Google StyleDrop generates images from text
https://techxplore.com/news/2023-06-google-styledrop-generates-images-text.html

StyleDrop uses Muse , which can generate images from text, and can generate images that adhere to a specific style. The feature is that by fine-tuning a small trainable parameter of less than 1% of the total model parameters, it efficiently learns new styles and improves quality through iterative training with feedback. It doesn't matter if you only have one image to base your style on.

Below is an example, the image in the orange frame on the left is the original image of the style, and the images on the right are images created based on that style. This is 'watercolor style'.



Van Gogh's 'Starry Night' style



Also Van Gogh's 'self-portrait' style



In addition to the realistic style, such a simple illustration style is also possible.



woodwork look



Not only can you render images, but you can also render text in a specific style. This is the 'Starry Night' style that was also there earlier.



Since foliage plants were drawn in the original illustration, foliage plants are placed on the left and right of the letters.



rainbow wavy alphabet



3D rendering style. 'C' has a doorknob.



You can also say, 'I made various things with Droid-kun's design.'



In addition, the developers said, ``We are aware of the potential pitfalls of being able to copy the style of individual artists without their consent, and we strongly demand responsible handling when using this technology.'' .

in Web Application, Posted by logc_nt