DramaChain Bench Pipelines 60 AI Short Dramas, Finds Upstream Defects Cost Up to 0.83 Points
Summary
DramaChain Bench, from Chinese researchers, evaluates every stage of AI short-drama production across 63 leaf dimensions and 5,785 items scored by three professional annotators. The paper shows upstream defects cascade — costing up to 0.83 points on multi-episode outputs — and the top pipeline (gpt-5.5-xhigh → gpt-image-2 → seedance-2.0) scores only 3.30/5.0, barely above the 3.0 usable line.
Originally reported by huggingface.co
Read the original article →Original headline: DramaChain Bench Pipelines 60 AI Short Dramas, Finds Upstream Defects Cost Up to 0.83 Points