English | 简体中文 | Demo | Discord

Album AI is an experimental project that uses the recently released gpt-4o-mini and Haiku as a visual model to automatically identify metadata from image files in the album. It then leverages RAG technology to enable conversations with the album.

It can be used as a traditional photo album or as an image knowledge base to assist LLM in content generation.

As a photography enthusiast facing terabytes of photos, I often felt overwhelmed. All existing photo management software required extra effort to maintain. Haiku and the newly released gpt-4o-mini gave me hope. So I decided to implement it immediately. My partner and I created the first version in less than 24 hours.

We hope you'll like it too. We welcome any praise or criticism. Don't forget to give us a ⭐️ or share to let more people know about it.

album.gcui.ai

• Automatically discover images in the album, using a PgSQL database for storage.
• Utilize GPT-4-o-mini to automatically generate metadata for images.
• Use OpenAI's Embedding API for metadata vectorization.
• Provide two APIs:
  • Search: A traditional Search API that takes a query and returns the most relevant images.
  • Chat: A RAG API that takes a query, retrieves images, and generates responses.
• One-click deployment to platforms like Render that support Docker container deployment.
• A permissive open-source license allows for integration and modification (please contact us for commercial use).

Recommended to run locally, if you want to run on a server, please deploy yourself, and we will improve this part of the guide.

git clone git@github.com:gcui-art/album-ai.git
cd album-ai

cp .env.prod.example .env.prod

Open .env.prod with your favorite editor, modify the configuration:

HOST_NAME= # Your local IP address, usually 192.168.x.x:8080
PROXY_URL= # (Optional) Your local proxy IP address, usually 192.168.x.x:7890, required when accessing OpenAI API directly is not available

OPENAI_API_KEY= # Your openai api key
ANTHROPIC_API_KEY= # Your Anthropic api key 

chmod a+x ./build.sh
./build.sh

Open the browser and visit http://localhost:8080 to see the demo.

Open the images directory in the project, add new photos to the images directory, and the background will automatically recognize and vectorize metadata. After that, you can use it in the demo through search and chat.

Album AI currently implements the following APIs:

• get /api/v1/file/search: Search for images
• post/api/v1/chat: Chat with images

There are four ways to support this project:

1. Fork the project and submit a PR: We welcome any PR to make Album AI better.
2. Submit an Issue: We welcome any reasonable suggestions or bug reports.
3. Recommend: Recommend the project to others; click Star; place a link to the project after using it.

Apache 2.0 License

We use Github's Issue to manage these feedbacks, you can submit one. We will often deal with them.

• Project repository: github.com/gcui-art/album-ai
• Album AI homepage: album.gcui.ai
• Live Demo: album.gcui.ai
• Author: @Kane

If you want to use it for commercial purposes, please contact us.