Promptist

Promptist is a research project from Microsoft that automatically rewrites a user's plain-language image description into a prompt better suited to Stable Diffusion. Instead of requiring people to learn the keyword-heavy style that diffusion models respond well to — adding modifiers for lighting, art style, or camera detail — Promptist takes a simple sentence and expands it into a more model-friendly version, aiming to improve the visual accuracy and coherence of the resulting image. The system was trained using reinforcement learning to optimize prompts specifically for how Stable Diffusion v1-4 interprets language, and the underlying method was accepted as a Spotlight presentation at NeurIPS 2023. It is distributed as an interactive demo hosted on Hugging Face Spaces and as open-source code and model weights on GitHub under Microsoft's LMOps project, so it is aimed at researchers, developers, and prompt engineers who want to study or reuse the technique rather than casual users looking for a full art-generation app on its own. Because the hosted demo runs on shared CPU infrastructure rather than a GPU, generation through the public Space can be slow, and free-tier Hugging Face Spaces like this one go to sleep after periods of inactivity, requiring a manual restart before the interface becomes responsive again.