GLEaN: A Text-to-image Bias Detection Approach for Public Comprehension

GLEaN is a research-oriented project aimed at solving a communication gap rather than a technical one. While most bias detection tools (like those from Stanford's HELM or various academic benchmarks) provide quantitative metrics, GLEaN focuses on 'public legibility' through visual portraits. With 0 stars and 3 forks at 7 days old, it is in the earliest stages of visibility. Its defensibility is very low (2) because the 'pipeline' is essentially a methodology for generating and arranging images to highlight stereotypes—a process that can be easily replicated by any developer with access to the same T2I APIs or weights. Frontier labs like OpenAI or Google are unlikely to build public-facing bias *exposure* tools for their own models, but platforms like Hugging Face could easily integrate these visualization techniques into their 'Model Cards' or evaluation leaderboards, posing a high platform risk. The primary value is pedagogical, making it a useful tool for journalists and researchers, but it lacks the 'data gravity' or technical complexity required for a higher defensibility score.