Text-to-Image Realism Evaluator
Completed image assessment and ranking preferences task for Outlier. The duties included analyzing pairs of images generated by an AI model using human-written text prompts and determining which image seemed more realistic and authentic. Criteria for assessing the images included realism, authenticity, and the presence of any smoothing effect, which is often found in AI-generated content. The purpose of this exercise was to train vision models to generate more realistic images that meet human expectations.