DATAREEL: Automated Data-Driven Video Story Generation with Animations

2026-04-29 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Expert, extended

Summary

DataReel introduces a new benchmark for automated data-driven video story generation, focusing on "data reels"—short, animated, chart-centric video clips with synchronized narration. This benchmark comprises 328 real-world stories, each pairing structured data, a chart visualization, and a narration transcript, enabling systematic evaluation of models in generating animated data video stories. The authors also propose a multi-agent framework that decomposes the task into planning, generation, and verification stages, mirroring human storytelling. Experiments show this multi-agent approach, utilizing models like Gemini 2.5 Pro, outperforms direct prompting baselines in both automatic and human evaluations, particularly in style consistency and overall visual quality. The benchmark and framework address the significant challenge of coordinating visual encoding, temporal progression, and narration, which typically requires substantial expertise in visualization design, animation, and video-editing tools.

Key takeaway

For research scientists developing automated data visualization tools, DataReel demonstrates that a multi-agent LLM framework significantly improves the quality and style consistency of animated data videos compared to single-prompt methods. You should consider adopting a multi-agent architecture for complex generation tasks, incorporating explicit planning and critical review stages to enhance output coherence and adherence to visual specifications. This approach can help overcome challenges in synchronizing animation, narration, and visual emphasis, leading to more effective data storytelling.

Key insights

DataReel benchmark and multi-agent framework automate animated data video generation, outperforming single-prompt LLMs.

Principles

• Decompose complex tasks into planning, generation, and verification stages.
• Synchronize animation, narration, and visual emphasis for effective storytelling.

Method

A multi-agent LLM framework uses a Director Agent for planning, a Plan Critic Agent for verification, a Coder Agent for HTML/D3.js generation, and a Video Critic Agent for rendered video evaluation and correction.

In practice

• Use D3.js for code-based animated visualizations.
• Employ LLM-as-a-judge for HTML code quality assessment.
• Prioritize style consistency in automated video generation.

Topics

• DataReel Benchmark
• Automated Video Story Generation
• Multi-Agent LLM Framework
• Animated Data Visualizations
• LLM Evaluation

Code references

• vis-nlp/DataReel

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Agentic Authoring of Interactive Multiview Visualizations in Genomics — Agentic LLM iteration significantly enhances genomics visualization quality, but complex agent architectures offer no further benefit.
• AI News: This Video Model Has Everyone Freaked Out! — New AI models are pushing boundaries in multimodal video generation, accelerated code synthesis, and cost-effective, agentic LLMs.
• Retrieval Augmented (Knowledge Graph), and Large Language Model-Driven Design Structure Matrix (DSM) Generation of Cyber-Physical Systems — LLMs, especially with RAG and GraphRAG, can automate Design Structure Matrix generation for complex systems.
• DSAEval: Evaluating Data Science Agents on a Wide Range of Real-World Data Science Problems — Multimodal, multi-query benchmarks are crucial for evaluating LLM-based data science agents on real-world, open-ended problems.
• MAVEN: A Multi-stage Agentic Annotation Pipeline for Video Reasoning Tasks — Maven's agentic, multi-stage pipeline generates structured video event annotations and CoT Q&A, enabling VLM domain adaptation and high performance.
• 🤖AI Agents Weekly: Claude Opus 4.6, GPT-5.3-Codex, Agent Primitives, METR Long Tasks, Codex App, OpenAI Frontier, C Compiler with Parallel Agents — Agentic workflows can automate end-to-end image generation and iterative refinement using LLM plugins.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.