Image Generation with Diffusion Model by Interactive Evolutionary Computation.

Text-to-image generation using deep learning based models has become a popular research topic, allowing users to generate custom artworks from specified text input. However, generating suitable prompts that produce creative and desirable images remains a significant challenge. To address this challenge, we propose a novel method that incorporates interactive evolutionary computation (IEC) to evolve the latent array. By integrating human perception into the system, our approach enables users to search for and generate images that align with their desired specifications through interaction with the system. We demonstrate the effectiveness of our proposed method in generating images that align with the users' mental images from an initial image using genetic algorithms through a series of experiments. Furthermore, the results from our user studies show that our proposed method enables users to generate images that match their desired mental images with less effort and in less time compared to conventional generation methods. Overall, this study contributes to the field of text-to-image generation by introducing a human-in-the-loop approach that enhances user control and specificity in the image generation process.