A Comparative Analysis of Different Versions of Stable Diffusion

Stable Diffusion has become a crucial technology in supporting the development of generative art, allowing users to generate detailed and diverse images from text prompts. Generative art is at the infant stage, but already attracted many researchers to develop the related technologies, which are quite diversified in terms of quality of images as well as algorithms and computing power that needed. Many versions of Stable Diffusion are available, but four versions will be selected in our study. Stable Diffusion-1.5 is relatively lightweight and requires low computing power, is good for quick prototyping and basic image generation. Stable Diffusion-2.1 with an improved CLIP text encoder for better prompt alignment and higher resolution. Stable Diffusion-XL supports 4K images and multi-modal input, aims at the market of professional and high-end needs. Stable Diffusion- 3.5 is a dynamic version, which uses super-resolution technologies to balance speed and quality. This study compares the above versions in terms of architectures, performance like generation speed and image quality, as well as application domains from digital art to scientific research. The results of this study can be used to support users choosing the best Stable Diffusion version for the tasks they have. Technologies that support Generative art is not just focus on the elements described in the prompt can be generated, but also the how the artistic components or art work can be generated that meet the user’s expectation.