DramaChain Bench: New Benchmark for End-to-End Short-Drama Generation

Haoyuan Shi (Hunyuan, Tencent), Mingtao Chen (Hunyuan, Tencent), Shuo Jiang (Hunyuan, Tencent), Ziyan Chen (Hunyuan, Tencent, Beijing Film Academy), Xuyi Sheng (Peking University), Yiming Liu (Hunyuan, Tencent), Ying Zhang (Hunyuan, Tencent), Miao Wang (Hunyuan, Tencent, Shenzhen University), Jianxiang Lu (Hunyuan, Tencent), Fanyang Lu (Hunyuan, Tencent), Songyuanyi Lu (Hunyuan, Tencent), Xiele Wu (Hunyuan, Tencent), Zhichao Hu (Hunyuan, Tencent), Yuhong Liu (Hunyuan, Tencent), Richeng Xuan (Hunyuan, Tencent)· September 2, 2026 View original

Key takeaways

  • DramaChain Bench offers the first end-to-end evaluation for short-drama generation.
  • It assesses all production stages, from script to final video, for consistency and intent.
  • The benchmark combines professional human annotation with an AI agentic judge for robust evaluation.
  • Upstream defects significantly impact final content quality, highlighting the need for holistic evaluation.

Who benefits

Media & EntertainmentAI/ML DevelopmentContent CreationAdvertising

Summary

DramaChain Bench is introduced as the first benchmark to evaluate all stages of short-drama production, from script to final video, addressing limitations of existing video-generation-only benchmarks. It uses a multi-dimensional evaluation system and professional human annotation, complemented by an agentic judge.

This paper introduces DramaChain Bench, a novel benchmark designed to evaluate the entire pipeline of short-drama production, moving beyond the typical focus on just video generation. The benchmark assesses each stage, including script, storyboard, keyframe imagery, and shot-level video, to ensure adherence to original script intent and coherence across assembled multi-episode releases. It leverages a comprehensive multi-dimensional evaluation system, DramaChain Dimensions, with 63 leaf dimensions, and incorporates both professional human annotators and an AI-powered agentic judge. The DramaChain Agent is calibrated against commercial short-drama platforms to ensure fair comparisons between models at each production stage. Human annotations, involving 5,785 items scored by three professionals, revealed that defects in earlier stages significantly impact final episode quality. An automated agentic judge, DramaChain Agentic Judge, was developed to score dimensions, achieving a high correlation with human rankings, which allows for cost-effective evaluation of new models.

Why it matters

This benchmark provides a standardized, comprehensive way for professionals in media production and AI development to evaluate and improve AI models across the entire creative pipeline, ensuring higher quality and consistency in AI-generated content.

How to implement this in your domain

  1. 1Adopt DramaChain Bench for evaluating internal AI-driven content creation tools.
  2. 2Integrate the benchmark's evaluation axes into your content quality assurance processes.
  3. 3Utilize the agentic judge for automated, cost-effective pre-screening of AI-generated short dramas.
  4. 4Analyze benchmark results to identify specific pipeline stages needing AI model improvements.

Original post by Haoyuan Shi (Hunyuan, Tencent), Mingtao Chen (Hunyuan, Tencent), Shuo Jiang (Hunyuan, Tencent), Ziyan Chen (Hunyuan, Tencent, Beijing Film Academy), Xuyi Sheng (Peking University), Yiming Liu (Hunyuan, Tencent), Ying Zhang (Hunyuan, Tencent), Miao Wang (Hunyuan, Tencent, Shenzhen University), Jianxiang Lu (Hunyuan, Tencent), Fanyang Lu (Hunyuan, Tencent), Songyuanyi Lu (Hunyuan, Tencent), Xiele Wu (Hunyuan, Tencent), Zhichao Hu (Hunyuan, Tencent), Yuhong Liu (Hunyuan, Tencent), Richeng Xuan (Hunyuan, Tencent)

"arXiv:2609.00646v1 Announce Type: new Abstract: Commercial short-drama production follows a multi-stage chain: script, storyboard, keyframe imagery, shot-level video, and the finished short drama. Most existing benchmarks evaluate solely the video-generation stage using pre-autho…"

View on X

Originally posted by Haoyuan Shi (Hunyuan, Tencent), Mingtao Chen (Hunyuan, Tencent), Shuo Jiang (Hunyuan, Tencent), Ziyan Chen (Hunyuan, Tencent, Beijing Film Academy), Xuyi Sheng (Peking University), Yiming Liu (Hunyuan, Tencent), Ying Zhang (Hunyuan, Tencent), Miao Wang (Hunyuan, Tencent, Shenzhen University), Jianxiang Lu (Hunyuan, Tencent), Fanyang Lu (Hunyuan, Tencent), Songyuanyi Lu (Hunyuan, Tencent), Xiele Wu (Hunyuan, Tencent), Zhichao Hu (Hunyuan, Tencent), Yuhong Liu (Hunyuan, Tencent), Richeng Xuan (Hunyuan, Tencent) on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses