Voxa - Your AI-powered comic finder

Search your comics with natural language and jump straight to the exact speech bubble — highlighted and read aloud.

Comment

🧠 Voxa – Your AI-Powered Comic Finder

🌟 Inspiration

I’ve always loved reading manhwa and comics, but over time I realized how difficult it is to remember or find specific moments among hundreds of chapters. There were times I wanted to revisit an emotional scene or a funny line — but scrolling endlessly wasn’t practical.

That inspired me to create Voxa, an AI tool that lets you search for any line or moment across your favorite comics and instantly highlight it right inside the image panel. It’s like giving your comics a searchable brain 🧩

💡 What it does

Voxa allows users to:

  • Search comic panels by text or natural language prompts.
  • Highlight exact speech bubbles or dialogues directly within the panel image.
  • Auto-scroll to the correct chapter and panel.
  • Enjoy an intuitive, AI-powered comic exploration experience.

The app uses hybrid search — combining both keyword and semantic understanding — so that even fuzzy or conversational queries (e.g. “when he confessed his feelings”) return relevant panels.

🏗️ How we built it

Frontend:

  • Built with Next.js (App Router) and React, styled using Tailwind CSS for clean and responsive design.
  • Implemented smooth auto-scroll and animated highlights using custom DOM overlays and keyframe animations.

Backend / Search:

  • Elastic Cloud for indexing and retrieving comic panels.
  • Google Cloud Vertex AI embeddings for semantic search vectors.
  • Indexed OCR-extracted text, panel metadata (bounding boxes), and base64-encoded panel images into Elastic.

Integration Flow:

  1. Comic panel → OCR text + bounding boxes
  2. Generate embeddings via Vertex AI
  3. Store in Elastic with hybrid retriever setup
  4. User search → Elastic query → Highlight exact region on panel

⚙️ Challenges we ran into

  • Elastic Cloud connection issues: For two days, I couldn’t connect because of Node.js version mismatches.
  • Trial limitations: Elastic’s 14-day trial made testing tight under hackathon time constraints.
  • Image coordinate mapping: Making bounding boxes align perfectly across different image resolutions was tricky.
  • Combining semantic + keyword search: Tuning weights between the two methods for best relevance took experimentation.

🏆 Accomplishments we’re proud of

  • Successfully built a working hybrid AI search that highlights text within comic images.
  • Learned how to connect Google Cloud Vertex AI embeddings to Elastic retrievers.
  • Deployed the full app live on Vercel — fast, stable, and simple to use.
  • Created an enjoyable user experience with instant visual feedback and auto-scroll.

📚 What we learned

  • How hybrid search combines lexical and vector similarity to improve context-aware retrieval.
  • How to visualize semantic search results directly inside images.
  • The importance of great UX when presenting AI results.
  • Working with real AI tools like Vertex AI, Elastic Cloud, and Next.js server actions for production-grade integration.

🚀 What’s next for Voxa

  • Build user libraries for saving favorite scenes.
  • Allow users to upload their own comics and auto-index them.
  • Use AI animation to bring still panels to life — turning comics into mini anime clips.
  • Eventually evolve Voxa into a full AI-powered webtoon and comic discovery platform.

Built With

  • elastic-cloud-(elser-semantic-text)
  • google-cloud-text-to-speech
  • google-cloud-vision
  • next.js
  • tailwind
  • vercel
Share this project:

Updates