Protect Art and Creativity: A Prevention Framework for Unauthorized Learning of Text to Image AIs.
In this work, we aim to solve the problem of unauthorized learning of works arising from mass data collection from Text to Image AI models, specifically Stable Diffusion. The TTI model performs indiscriminate web data crawling to collect images, and these images are used for model learning without the consent of the author. The TTI model is capable of learning the drawing style of an image, which undermines the value of the original work. Therefore, we suggest a method of transforming images to deteriorate the learning accuracy of TTI models and evaluate the quality of the generated image using both quantitative measurement and qualitative measurements. Thus, we confirm that our proposed image modification method prevents AI models from learning literary works without permission.
Share this Project