Controllable Image Generation and Artistic Style Transfer

In recent years, with the rapid development of diffusion models, a variety of image generation and editing algorithms have emerged. One significant topic in this field is controllable generation. Within this area, a particularly artistic research problem is style transfer, which aims to apply the style of one image to another while preserving the content structure of the latter.

This project focuses on generating images in a specified style. Given a series of reference images sharing the same style, along with a designated prompt, we train models to guide the generation of images in the same style.

Our approach requires separate weight fine-tuning for each style. For each style, 10 images are used for training. By inputting a prompt describing a new object, the model generates images of the object in the desired style.

Report: Coming Soon

Link: GitHub Repository