Art Generation AI Model for Low-End Devices

Art is a powerful medium of emotional expression, using elements like color, sound, and form. It reflects the artist's personality and the era they lived in, telling captivating stories. AI has advanced in generating and manipulating high-quality images using text prompts. Generative Adversarial Networks (GANs) revolutionized image generation, leading to various models like Open-Edit, GLIDE, DALL-E, and VQGAN. These models have shown impressive capabilities, producing creative outputs. However, accessibility remains a challenge due to the computational resources required.This work aims to enable art generation on low-end devices, making AI-driven art more accessible. One major challenge identified in the models is the high memory demand during image creation. Here, the chosen method, VQGAN-Clip, allows for image creation and manipulation solely through human-written text prompts. To address the memory constraint, the sliding window method is used in conjunction with VQGAN-Clip. This approach effectively reduces the VRAM requirement from 8 GB to 2 GB for generating images of the same size.