ABSTRACT
With the advancements in artificial intelligence, its application for image generation has become so much more natural. However, using just one particular technology such as ControlNet, IP Adapter, Image Inpainting or GAN-model restricts us from generating very realistic images which makes the application of these models difficult. Hence, in this research, we describe ICIP (Image inpainting, ControlNet and Ip adapter) a new integrated approach to fine-tuning stable diffusion models using ControlNet, IP Adapter, and image inpainting technologies together, which generates user-specified, accurate, and adaptable images. We found ways to constrain the model to produce strikingly informative and beautiful images in the appropriate styles by fine-tuning a custom dataset. Initially, we propose a text-to-image generation method, which provides realistic images using user-specific prompts, subsequently, we can perform image-to-image enhancements and image inpainting technique which allows us to mark the region of interest from the output image and provide a prompt to make desired changes and refine the generated image to produce an accurate and realistic output image. Our proposed model generates clearer and more authentic images than the existing Diffusion Models, leading to impressive and adaptable image formations.
