Mozart

Inspiration

Our inspiration comes from several places - talks by Bret Victor (worrydream), watching masterful graphic designers communicate and execute, as well as the recent advances in (not quite composable) text to image generation. We believe that good design and copy should be accessible to everyone with an idea.

What it does

Our AI design assistant helps you design posters and write good inline copy for your graphic design projects. This could be anything from an invite to your birthday bash to a menu for a steak restaurant. The user can guide the AI via UI/UX prompts as well as natural language prompts about the nature of the project.

The three step workflow to build a poster:

Pick among the given workflows (event, contest, menu, wedding)
Choose a base template that fits your project
The AI will generate a poster for you, which you can use or edit

How we built it

This was a technically challenging project for the scope of a hackathon. At a high level, we built :

A front-end app that can help you design, edit content and manipulate objects on your browser. The tech stack here is vanilla JS and JQuery for simplicity and flexibility.
A back-end that processes generation requests from the front-end by calling OpenAI's davinci models and parsing the response. The tech stack here is Python Flask.

Challenges we ran into

We spent a lot of time trying to solve the chicken-and-egg problem of figuring out what prompts have hi-quality responses, and what UI/UX to build around those prompts.

Even though we have ML backgrounds, "prompt engineering" is brand-new to us. I never thought it's a sustainable/effective way of leveraging LLMs. But turns out, LLMs can be coaxed into becoming powerful oracles from stochastic parrots with the right inputs.

After a large number of iterations, we settled on workflows that looked magical once everything was tied together.