Turn structured data into narrated, animated data stories.
For analysts, researchers, and anyone who wants their data to tell a story.
🔥 News • ⚡ Quick Start • 🌟 Examples • 🎯 Workflows • 📖 Citation • 🤝 Community
Try it: https://datamagic.chat/
Project homepage: https://datamagic-home.github.io
Paper: DataMagic: Transforming Tabular Data into Data Insight Video
- [2026.07.05] ✨ Added a Customizable Generation Workflow: DataMagic now surfaces recommended scene plans, visual designs, narrative ordering, and animation highlights before final rendering, so users can review and adjust key decisions instead of editing only after generation.
- [2026.06.20] 🚀 DataMagic is now live! Try it at datamagic.chat — upload your data and generate a narrated data video in minutes.
- [2026.06.20] 🧩 Released the datamagic-video skill — reusable guidance that teaches AI coding agents (Claude Code, Cursor, Codex) to turn tabular data into narrated data videos.
- [2026.06.18] 📄 Our paper "DataMagic: Transforming Tabular Data into Data Insight Video" has been accepted to VLDB 2026 Demo Track and is now available on arXiv.
Most teams already have tables. The hard part is turning those tables into something other people can quickly understand: finding what is worth saying, choosing the right charts, arranging the story, writing narration, timing animations, and producing a video people can actually watch.
DataMagic is built to remove that repetitive work. It helps analyze the data, surface useful insights, and turn them into an editable narrated video you can preview, refine, export, and share.
Today's tools are strong in their own domains: Excel, Vega-Lite, and Matplotlib are great for charts; Tableau, Power BI, and Looker are strong for exploration and monitoring; After Effects, Premiere, and CapCut are good for video editing; Seedance, Sora, and Veo can generate visual footage. But going from a raw table to a playable, editable, and traceable data-story video still requires analysis, chart selection, narration, animation timing, and data checking.
DataMagic focuses on this missing workflow: turning raw structured data into an editable, traceable, narrated data video.
DataMagic is an AI-assisted system for authoring data videos from tabular data. Upload a CSV or Excel table, provide an analysis goal or business question, and DataMagic helps analyze the data, surface insights, plan the story, choose charts, draft narration, synchronize animation, preview the result, and export an MP4 video.
The goal is not just to "generate a video." The numbers, labels, and charts in a DataMagic video should remain connected to the original table. Under the hood, DVSpec connects visual elements, narration, and animation timing so the result is easier to inspect, edit, and extend.
| What you want to do | Where common tools fall short | How DataMagic helps |
|---|---|---|
| Turn a table into a video | Analysis, charting, scripting, and editing usually happen in separate tools | One workflow from uploaded data to narrated video |
| Let AI find what matters | Many tools draw charts but do not decide what is worth saying | Analyze the data and organize useful findings into scenes |
| Keep numbers and charts verifiable | Pixel-level video models primarily generate frames; data binding and provenance require extra verification | Bind chart elements to source data through DVSpec |
| Edit after generation | Regeneration or manual video editing is often required | Preview, edit text, refine with natural language, and keep changes local where possible |
| Share the result quickly | Static charts still need someone to explain them | Export a playable animated data story |
- Upload your data — CSV or Excel table
- Review and adjust DataMagic recommendations — scene plans, chart designs, story order, templates, and animation highlights
- Export your video — download the finished data story
Full Pipeline — Starting from a data table, AI automatically analyzes the data, plans the narrative structure, and generates narration and animations for each scene, with animations synchronized to the narration automatically. The result is a complete multi-scene data video. Best for high-quality presentations such as business reports, research showcases, and sharing analytical conclusions with your team or leadership.
Fast Generation — Follows the same process as Full Pipeline — AI still handles content planning and narration — but scene rendering uses pre-built visual templates instead of per-scene AI generation, making it significantly faster with less visual customization. Best for users who prioritize speed, such as recurring report production or when visual style is not a primary concern.
Single Chart — No full video needed, just one animated chart to discover or explain a focused data point. Paste your data and quickly generate a single animated chart, ready to embed in a presentation, report, or social media post. Best for quick exploration or communicating a local insight without a full narrative structure.
Note
Fast Generation and Single Chart are experimental features currently in beta. They work well for typical inputs but may produce unexpected results in edge cases. Feedback and bug reports via Issues are very welcome.
Both Full Pipeline and Fast Generation support a customizable mode. DataMagic surfaces its recommendations at key stages, including scene planning, visual candidates, narrative ordering, visual templates, and animation highlights. Users can accept the recommendations or adjust scenes, charts, templates, and story order before final rendering. Fully automatic generation remains available for users who want the fastest end-to-end result.
DataMagic-Demo.mp4
System walkthrough — from data upload to narrated animated video.
China-Consumption.mp4China consumption recovery Analyzes China's retail sales and catering revenue from 2019 to 2025, showing consumption resilience and service-sector recovery after the pandemic shock. |
China-EV.mp4China EV market competition Compares 2024 monthly sales, annual pacing, and year-over-year growth across major EV brands, highlighting market leaders and growth inflection points. |
Q4-Sales.mp4Q4 sales analysis Animated bar and trend visualization for business performance insights. |
Renewable-Energy.mp4Renewable energy transition A narrated look at global renewable capacity growth from 2018 to 2024, highlighting solar expansion and the declining share of fossil fuels. |
Tech-Revenue.mp42024 tech revenue leaders A comparison of major technology companies by 2024 revenue, showing Amazon's scale alongside Apple, Google, Nvidia, and Meta. |
Tech-Growth.mp4Tech growth and market momentum An executive-style recap of 2024 tech performance, contrasting revenue scale with fast growth led by Nvidia. |
Over 100 ready-made visual styles across bar, line, pie, scatter, Sankey, waterfall, KPI card, and more — each with a preview and community ratings. Browse and mark your preferred styles before generation.
DataMagic is built around two core principles: data-grounded scenes (every visual element bound directly to a data field, keeping the story fully traceable and editable) and narration-aware timing (animations auto-synced with the voiceover, producing a coherent narrative rather than a collection of disconnected charts).
- AI-assisted chart type and visual template recommendation.
- Customizable generation workflow for reviewing and adjusting DataMagic recommendations during scene planning, visual design, narrative arrangement, and animation highlighting.
- Runtime preview, direct visual editing, and natural-language refinement.
We also publish datamagic-video — a skill that teaches AI coding
agents (Claude Code, Cursor, Codex, …) the methodology behind data videos: narrative
patterns, chart selection, DVSpec authoring, narration writing, and animation timing. The
videos it produces render with open tooling, so anyone can generate and watch them — no
account needed.
The hosted product adds premium templates and the full pipeline; the skill gives any agent strong standalone results and shares the same DVSpec format.
claude plugin marketplace add HKUSTDial/DataMagic
claude plugin install datamagic-video@datamagicCodex:
codex plugin marketplace add HKUSTDial/DataMagic
codex plugin add datamagic-video@datamagicOr install from shell:
curl -fsSL https://raw.githubusercontent.com/HKUSTDial/DataMagic/main/install.sh | bashThen ask your agent: "Make a narrated data video from this CSV …". See the skill README for details.
The source code is being progressively open-sourced — ⭐ star this repo to follow updates.
|
Join the WeChat discussion group Share use cases, ask questions, and discuss data visualization or AI-generated videos with the community. |
 Scan to join |
Follow either official account below and reply DataMagic to receive a one-time credit code (20 credits) for the hosted product.
 DIAL 实验室 Research updates & DataMagic news |
 蟹哥聊科研 Research & AI tool tips |
- Core generation modes — Full Pipeline, Fast Generation, and Single Chart.
- Template gallery and runtime editing — preview styles, edit generated text, and refine with natural language.
- Bilingual public documentation — English and Chinese release docs.
- Data-video skill package — reusable guidance for data-video planning, chart selection, DVSpec authoring, and animation design. (skills/datamagic-video/)
- More diverse visual styles — richer narrative cards, report themes, domain-specific templates, and presentation-ready layouts.
- Recommendation and feedback learning — improve template ranking from user preferences and real generation outcomes.
- Public implementation materials — clearer notes for the pipeline, DVSpec, template adapters, example datasets, and deployment.
- Expanded export and sharing workflows.
- Team/admin monitoring for production deployments.
If you find DataMagic useful in your research or work, please cite:
@misc{xie2026datamagictransformingtabulardata,
title={DataMagic: Transforming Tabular Data into Data Insight Video},
author={Yupeng Xie and Chen Ma and Zhenyang Wang and Liangwei Wang and Jiayi Zhu and Chuxuan Zeng and Zhouan Shen and Boyan Li and Yuyu Luo},
year={2026},
eprint={2606.20388},
archivePrefix={arXiv},
primaryClass={cs.HC},
url={https://arxiv.org/abs/2606.20388},
}- Data-Video Skill
- Pipeline Overview
- DVSpec Overview
- Input and Output Examples
- Help Center
- Release Status
