How it works
Infinite Wonderland is made possible by the artists collaborating with Google Deepmind’s AI models
Gemini + Imagen
01 Each Artist Created
Original Images for the Story
Inspired by John Tenniel’s original illustrations, each artist created a small bespoke set of images visualizing the novel through their own eyes. They wrote descriptions for each image and defined their own unique style.
02 The Artists Fine-Tuned Imagen 2 on the Style of Their Original Images
Using a fine-tuning technique called StyleDrop, the artists fine-tuned an image generation model, Imagen 2, on their styles using a handful of their original images. This was an iterative process where each artist could see how their original images impacted the model’s outputs and then make creative alterations until the model generated the aesthetic and composition that felt most authentic to their style. Once complete, these fine-tunings allowed them to generate images of anything they described in their own unique styles.
03 Every Sentence of the book is transformed into a custom image prompt using Gemini
There are over 1200 sentences in Lewis Carrol’s original novel. Using a few-shot prompt, Gemini transformed each sentence into an image description. Then, each image description was customized to each artist’s fine-tuned style as a prompt for Imagen 2.
04 Each Sentence Can Generate Infinite Images in Any of the Artist’s Fine-Tuned Styles
By combining each artist’s fine-tuned style with their custom image prompts, each sentence of the book can generate seemingly infinite images in any of the artist’s fine-tuned styles. This combination is the core of the Infinite Wonderland experience that enables this timeless classic to be endlessly reimagined by artists, AI and you.