Google released a neural network StyleDrop, it can create images in a precisely defined style

StyleDrop learns the style of any image and helps a generative artificial intelligence model recreate it. Google's method is superior to others, such as Dreambooth, LoRA, or Textual Inversion.

Google's new method lets you synthesize images in a specific style using Muse's text-image model. StyleDrop conveys all the subtleties of custom styling, including color schemes, shading, design patterns, local and global effects. According to Google, only one image is required as input.

StyleDrop learns a new style by fine-tuning a small number of parameters of the training network, and then improves the quality of the model by iterative learning with human or automatic feedback.

StyleDrop learns quickly and on a small number of examples

Specifically, StyleDrop is trained on an input image and generates a set of images to play back that image. From these, the highest quality images are selected, either by CLIP score or by human feedback, and used for further training. An image is considered high quality if it reproduces not the content but the style of the original image.

According to the team, the entire process takes less than three minutes, even with human feedback. That's because StyleDrop requires less than a dozen images for iterative learning, they say.

StyleDrop is superior to other methods of transferring style from text to image, including Dreambooth, LoRAs, Textual Inversion in Imagen and Stable Diffusion, according to the team.

StyleDrop for styles, Dreambooth for objects

"We see that StyleDrop is able to capture nuances of texture, shading and structure across a wide range of styles, much better than previous approaches, allowing much more control over style than was previously possible," the team said.

The team also combined StyleDrop with Dreambooth to explore and create a new object in different styles as an image and can use methods with Muse to create a custom object in a custom style.

Google sees StyleDrop as a versatile tool, one use of which is for designers or companies to practice with their brand's resources and quickly prototype new ideas in their style. More information can be found on the project page StyleDrop.