What if a regular knitted toy had its own travel photo diary?
In this project, I trained the generative model Stable Diffusion XL on images of a pink knitted rabbit to create new photographs featuring the same character in various locations around the world. London, Paris, Amsterdam, and other cities serve as the backdrop for its imagined journey. My objective is to test whether the neural network can maintain the toy’s recognizability, its vibrant color, and the texture of the knit while altering the environment, composition, and narrative.
Examples of source images
From Toy to Digital Character
I knitted this pink bunny myself, and now I take him with me on every move. But what if he doesn’t just accompany me—but starts traveling the world on his own?
How I Taught My Bunny
For this project, I utilized Stable Diffusion XL and LoRA—a technique for fine-tuning a neural network using my own images. I gathered photographs of the pink rabbit I created and used them to train the model. The main goal was to teach the neural network to reproduce my specific character, preserving its vibrant color, knitted texture, ear shape, and recognizable features. After the training, I began generating new scenes, altering the cities, environment, and composition. Thus, a simple toy became the protagonist in a digital travel photo diary.
How the neural network was trained
For this project, I utilized Stable Diffusion XL combined with the LoRA method. This allows one to fine-tune the model on a specific character without training the entire neural network from scratch. I used photographs of my own knitted bunny as the foundation. My objective was to teach the model to recognize its distinct features and reproduce them within new scenarios.
Stages of Work
code snippet
