AI Photo Of Me is an innovative app that allows users to upload 10-20 photos of themselves and generate custom profile photos in up to 100 styles.
The app is an implementation of Stable Diffusion 1.5 combined with DreamBooth (to train faces) and RealESRGAN (to upscale to 4k).
ROLE
PERSONAL PROJECT (AKA EVERYTHING)
TOOLS
PROBLEM
Lensa AI was becoming increasingly popular and I wanted to create a similar app that was more customizable and cheaper than Lensa AI. So I decided to create AI Photo Of Me.
Using this Google Colab I was able to learn how to train faces using Stable Diffusion 1.5 and DreamBooth. Using python and pm2 I created a script that would check a database for new entries and train them.
The next step was to create a frontend for the app. I decided to use Next.js and Tailwind CSS to create the frontend. I also used Framer Motion to create the animations. Stripe was used for payments and PostgreSQL for the backend.
After several iterations of the frontend, I was able to create something that I was happy with. I also created a backend using PostgreSQL and Next.js API Routes to handle payments and user data.
The next step was to upscale the images to 4k. I used RealESRGAN to do this. By adding the 4k option only to the premium plan, I was able to increase premium plan sales from 3% to a whopping 50%!
SUMMARY
AI Photo Of Me is a personal project that I created to learn more about AI and image processing. I was able to learn how to train faces using Stable Diffusion 1.5 and DreamBooth.
Using simple free marketing techniques, I was able to gain 800+ daily users with a best sales week of $123 (I know it's not much but it's a start!)
Marko Pekas © 2023