Pretty Bench

Pretty Bench provides a public, human-preference benchmark for evaluating generated creative work, including images, video, and text. Key features include: • Head-to-head model comparisons • Real human voting to collect preferences • Transparent rankings and leaderboards • Exploration of model outputs • Focus on style, taste, and preference This platform allows users to put various generative systems to the test, gathering direct human feedback to determine which creations people genuinely prefer. By engaging a community of voters, Pretty Bench moves beyond purely technical scores to build a more nuanced and useful picture of output quality. Participants can explore detailed rankings, compare the results of different models side-by-side, and contribute their own votes to shape the leaderboards. This collaborative approach helps to refine the understanding of what makes generated content compelling and valuable to human observers. It offers dedicated arenas for image, video, and text evaluation, ensuring a comprehensive assessment across different creative domains. Pretty Bench is ideal for researchers, developers, artists, and content creators seeking to understand real-world human perception of generated content. It serves as a crucial resource for anyone interested in gauging the effectiveness and appeal of different generative models from a user-centric perspective, fostering advancements that align with human taste.