ShareGen AI
Inspiration
A year and a half ago, after graduating from UC Berkeley, I started a small business with just two people. We resell used lab equipment from schools and hospitals so it doesn't go to waste and can be used again by underfunded schools, students, or people who want to do research or build things but can't afford to spend $10,000+ on a new oscilloscope, vacuum pump, or centrifuge.
The company is doing well, but the biggest challenge has always been marketing and finding customers for such a niche, small business, a problem that almost every small business owner faces. It is always time-consuming and expensive to hold photoshoots, create animations, generate videos, and run campaigns - factors that are not core to our business, but are vital to our success.
In today's world of AI video generation, this process should be completely streamlined. Ad campaigns should be automatically run by AI agents that produce content on your behalf, optimize it for social media and search engines, and run both paid and organic ads. Small businesses have great products and solve real problems, but may not always have the scale or expertise to make themselves visible to their customers. The technology already exists to automate this, it just needs to be packaged in a way that actually works.
What it does
ShareGen AI automates the entire advertisement creation process for small businesses. It goes from product photos or Shopify links to custom video advertisements (image posts coming soon) by leveraging state-of-the-art video generation, AI music creation, and some clever post processing techniques.
How we built it
The application takes in a Shopify link or custom images submitted by the user, then collects product images from the Shopify store and presents them to the user while sending off a task to GPT-4o to analyze the images and find the best ones to be used as advertising material.
Once the user is satisfied with the images for their advertisement, we send off the images to MiniMax for state-of-the-art video generation (up to 10 seconds per clip and 1024P). We use a music generation API to generate a custom song for the advertisement that matches the brand's themes, and we find publicly available, free-for-commercial-use stock footage based on the brand's advertisement plan (all decided by AI to fit each brand).
Finally, we use our custom processing algorithm to select, edit, and convert all these newly created or found video, image, and audio assets into one coherent and stunning advertisement (up to 60 seconds long!).
Challenges we ran into
Unfortunately, the speed of state-of-the-art video generation in 2025 is still painfully slow, with some clips taking up to 8-10 minutes per generation. Similar struggles came up with music generation and simple image processing. These long processing times created significant user experience challenges.
Managing the complex pipeline of multiple AI services, each with their own processing times and potential failure points, required careful orchestration and error handling. Coordinating the timing of video generation, music creation, stock footage selection, and final asset compilation while maintaining quality standards proved to be a significant technical challenge.
Accomplishments that we're proud of
We optimized our platform to run dozens of tasks in parallel and quickly process media assets to deliver the best and fastest experience for our users, something we are extremely proud of. Despite the inherent slowness and hallucinations of current AI generation technology, we've managed to create a system that maximizes efficiency through intelligent task scheduling, parallel processing, and custom data processing.
We successfully created a complete end-to-end solution that transforms a simple Shopify link into a professional, 60-second advertisement with custom video, music, and carefully selected imagery. The platform demonstrates how AI can democratize professional advertising creation for small businesses that previously couldn't afford such services.
What we learned
Working on this project taught me about some powerful tools I hadn't fully appreciated before. I'd tried Bolt.new a few times previously but didn't realize how powerful it was at generating fantastic UIs and mapping data around an application. I also rarely used Supabase before this project, but I found the Supabase edge functions super useful compared to running Postgres on AWS.
One thing I'd never used before was Postgres cron jobs, which I implemented to set up a job orchestrator that checks my video generation database every minute to see if very long API tasks (up to 10 minutes) have been completed and progresses them to the next step of the video processing pipeline.
What's next for ShareGen AI
Future plans include expanding beyond Shopify integration to support other e-commerce platforms, improving the speed and quality of generated content, and adding more sophisticated targeting and optimization features for ad campaigns across different social media platforms.
We're also developing an advertising agent that will be able to automatically generate advertisements on behalf of users and queue them for scheduled uploads to ad campaigns or organic social media. This means a business's entire ad department can potentially run on our platform, providing complete automation of their marketing efforts.
Built With
- bolt.new
- minimax
- openai
- stripe
- supabase
- typescript
Log in or sign up for Devpost to join the conversation.