The idea for this sparked when I saw friends dropping a fortune on engagement photos, right around the time the Taylor Swift engagement pictures went viral. I knew there had to be a better way.
The final push came recently with all the hype around Google's Banana model. I started playing with it via the Gemini API and the quality was just incredible, so I decided to build this SaaS to generate engagement photos.
This was my first crack at building an app like this, and the main hurdle quickly became prompt engineering. It all boiled down to: how do you consistently guide the model to produce the right image? After a ton of trial and error, I landed on a prompting formula that works. I even put an LLM in the middle of the workflow to refine the prompts, which honestly feels like I'm building a primitive agent.
Anyway, here it is. You get 5 free generations to try it out. I'd genuinely love any comments or feedback you have. Thanks for checking it out.
maxbaines•5mo ago
I am very much pro AI and use it daily for everything from code to product photos, and games. But I feel these moments are one of the few you actually want to keep AI away from. Imagine looking back in hopefully 50years at an AI image.
michaellzd0303•5mo ago
Yes, that is a great point. Real moments and memories are irreplaceable. We don’t see AI as a replacement for traditional photos, but as an option for couples who may not have the budget, time, or resources for a professional shoot. We just want to have another option for people who would like to use AI to create some funny and low-cost photos to share in their social media( Funny but filled with love!)
michaellzd0303•5mo ago
The final push came recently with all the hype around Google's Banana model. I started playing with it via the Gemini API and the quality was just incredible, so I decided to build this SaaS to generate engagement photos.
This was my first crack at building an app like this, and the main hurdle quickly became prompt engineering. It all boiled down to: how do you consistently guide the model to produce the right image? After a ton of trial and error, I landed on a prompting formula that works. I even put an LLM in the middle of the workflow to refine the prompts, which honestly feels like I'm building a primitive agent.
Anyway, here it is. You get 5 free generations to try it out. I'd genuinely love any comments or feedback you have. Thanks for checking it out.