art-rethinking-ai

This project uses OpenAI's APIs to recreate famous paintings through an image-to-text-to-image process.

The goal is to evaluate how well AI can analyze and recreate paintings without explicitly naming the original title or artist.

By providing an image of a famous painting (e.g., the Mona Lisa), OpenAI's APIs generate a detailed description of the artwork, including the objects and colors depicted.

Using this description, the project then prompts the OpenAI APIs to produce a new, original painting.

The title and artist of the original masterpiece are intentionally omitted.

Examples

These are some examples of artworks we have created with this project:

ORIGINAL	NEW

Usage

You can try this project locally.

Requirements

• Node >= 18: you can download it here or using a Node Version Manager;
• an OpenAI's account: you can create it here;
• at least 5$ in your OpenAI's wallet: you can charge it here;
• an OpenAI API key: you can create it here;

Get started

Once cloned this repository, install its dependencies:

npm i

Create a file called .env.local in the root folder of the repository, and add your OpenAI API key here:

NEXT_PUBLIC_OPENAI_API_KEY=(openai-api-key-here)

Now you can run the project

npm run dev

And test it on http://localhost:3000.

Contribute

To contribute to this project, you can follow the instructions reported in Usage above.

The technologies used in this projects:

• OpenAI APIs;
• React;
• Next.js by Vercel;
• SWR;
• Materialize CSS;
• WikiArt for paitining images and data;

TODO

• Add the possibility to search a painting (e.g. using WikiArt APIs);
• improve the UI components (migrate to Bulma?).