Skip to content
3rd Place

Art

Art is presented as static objects with text-heavy placards and audio guides, while the real reasons people care: conflict, emotion, symbolism, and the artist’s lived drama, are compressed into a few lines of static text

Code Video Cerebral Valley

Core trick

Photo goes to multimodal Gemini for art-history research, then narration and generated visuals are assembled into a 4-scene video.

Tags

creative/musiceducation

Description

Art is presented as static objects with text-heavy placards and audio guides, while the real reasons people care: conflict, emotion, symbolism, and the artist’s lived drama, are compressed into a few lines of static text. Meanwhile, millions of letters, diaries, and historical records remain “data graveyards,” effectively locked away from the public. ArtLens AI redefines how people experience and feel art. We’ve built a new way to consume artwork: point your phone at any piece and it instantly becomes a short, cinematic micro-documentary that brings the hidden story behind the canvas to life in a format native to a digital generation. We redefine the boundary between audience and creator by turning the act of looking at art into the start of an interactive story. Under the hood, we capture a photo and use Google AI tools for multimodal understanding and generation and narration and visuals into a finished film, centered on the original artwork.
djvaroli/artlens-ai★ 20 forksTypeScriptMITCreated Feb 28, 2026 · last push Mar 5, 2026

README excerpt

<p align="center">
  <img src="assets/artlens-ai-logo.jpg" alt="ArtLens AI" width="160" />
</p>

<h1 align="center">ArtLens AI</h1>

<p align="center">
  <strong>Upload any artwork. Get a cinematic video documentary in minutes.</strong>
</p>

<p align="center">
  🏆 3rd Place at the Google DeepMind NYC Hackathon 🏆
</p>

Team members: 

Daniel - https://github.com/djvaroli/

Mike - https://github.com/mikeeekoch-cmd/


## Demo

Here's a generated documentary for Salvador Dalí's *The Persistence of Memory*:

https://github.com/user-attachments/assets/e27aa07f-48b9-49a8-9397-d35958127fd6

## What It Does

ArtLens takes a single image of any artwork and turns it into a 4-scene video documentary with AI-researched art history, cinematic narration, and smooth scene transitions. The whole pipeline runs on Google's generative AI models.

```
Upload Artwork → Research (Gemini + Search) → Scene Scripts (Gemini Flash) → Video (Veo 3.1) → Stitch (ffmpeg)
```

### The Pipeline

1. **Research** - Gemini Flash analyzes the image and runs grounded Google Search to build a structured dossier: cinematic hooks, primary source quotes, symbolism, historical context.
2. **Scene Scripts** - Generates a 4-scene arc (Introduction, Technical Detail, Emotional Context, Closing Reflection) with title cards, visual descriptions, and narration.
3. **Visual Prompts** - Translates each scene into a Veo-optimized prompt with consistent framing: static camera, artwork centered, no fades.
4. **Video Generation**

Team