At Google I/O, Google introduced a product called "Infinite Wonderland," which is similar to the one shared earlier.
Experience
On the official website https://infinitewonderland.withgoogle.com/, you can select an artist theme you like and then choose a corresponding piece of text to generate illustrations for the story.
Through this method, users can personally experience the creative charm of Infinite Wonderland, perfectly combining classic novels with modern technology to create unique visual stories. Each generated image may be one-of-a-kind, showcasing the infinite possibilities of collaboration between artists and AI.
Technical foundation
and
is Google's highest-quality text-to-image generation model, capable of producing detailed images with rich lighting effects and minimal noise artifacts. It understands natural language prompts and can generate a variety of visual styles while capturing the details in complex prompts.
allows users to generate high-quality images that match a specific style by providing a single reference image. Through efficient parameter fine-tuning and iterative training, it achieves meticulous capture and reproduction of the style provided by the user.
Implementation process
01 Each artist creates original images for the story.
Inspired by John Tenniel's original illustrations, each artist created a small set of custom images to reinterpret the novel through their own lens. They wrote descriptions for each image and defined their unique styles.
02 Artists fine-tune Imagen 2 based on their original artwork style.
Using a fine-tuning technique called StyleDrop, the artists fine-tuned the image generation model Imagen 2 with their original artworks. This was an iterative process where each artist could see how their original artwork influenced the model’s output and then make creative adjustments until the aesthetics and composition generated by the model felt most aligned with their style. Once completed, these fine-tunings enabled them to generate images of any description in their unique style.
03 Every sentence is turned into a custom image prompt by Gemini.
The original novel by Lewis Carrol contains over 1,200 sentences. Using a few example prompts, Gemini transformed each sentence into an image description. Then, each image description was customized according to the fine-tuned styles of individual artists as prompts for Imagen 2.
04 Every sentence can generate infinite images in any artist's fine-tuned style.
By combining the fine-tuned styles of each artist with their custom image prompts, every sentence in this book can generate what seems like an infinite number of images in any artist's fine-tuned style. This combination is at the heart of the Infinite Wonderland experience, allowing this timeless classic to be endlessly reimagined through artists, AI, and users.