Cubespace

Our content review page
An example of a Tweet generated by Cubespace
Another example of a Tweet generated by Cubespace
Our main page

Inspiration

Web marketing is an increasing part of our daily life, and it's an aspect of marketing small business owners especially can't ignore. However, it's difficult for small business owners (who might not have the resources to hire dedicated social media professionals) to compete with large corporations. To even the odds, the Cubespace team wanted to find a way to harness the power of AI to help create and deploy product marketing on one of the world's largest and most cost effective advertising platforms: Twitter. The result? Cubespace, the next dimension for online marketing.

What it does

Cubespace leverages the power of AI by using OpenAI's GPT-3 and Dall-E deep learning models to dynamically and auto-magically generate and deploy marketing content. Users simply describe their product and a marketing image using natural language, and Cubespace is able to handle create marketing pitches and graphics, and deploy them on to their Twitter account.

How we built it

We built Cubespace using the OpenAI APIs for GPT-3 and for Dall-E. To properly create tweets that would fit the style and character limit of Twitter adverts, we fed the generative text model Twitter ads from other companies. We also engaged in prompt engineering for the image generation, cycling through dozens of potential art styles and prompts before settling on our minimalist and flat design. The text and images generated would be posted using the Twitter API, and we used Flask, HTML5, and CSS to create a web application and front-end for users to interact with.

Challenges we ran into

A major challenge we had was in prompt engineering. We quickly discovered that Dall-E images generated in a realistic or photorealistic manner was incredibly disturbing and grotesque, and ultimately not suitable for marketing materials. We also discovered limits in trying to get tweet content that was both dynamic and informative, but also fell within the 280 character limit of Twitter. Finally, we experienced lots of difficulties in trying to use Flask to integrate our backend and front-end.

Accomplishments that we're proud of

We're happy that we finished the app and powered through all the difficulties. We're also happy that we were all able to enter the hackathon and leave with a new skill: none of us had any front-end experience at all before this hackathon. We're happy to finish the project and leave with a cool new skill.

What we learned

How to create and deploy a web app! How to do sentiment analysis. Oauth1 and 2 tokens. The joys and pains of having to read horribly vague documentation!