I Spent 5 Days Building My First SaaS — Here's What I Learned

You might not believe this, but I haven't had a "real job" in 5 years — and I've been low-income the entire time. During this time, I've tried many projects — mobile apps, freelancing gigs, remote work — but none of them worked out well.

Recently, AI technology has advanced rapidly, with impressive new AI products emerging every day. This has created many opportunities for entrepreneurs like me. At the same time, AI-powered coding tools have become so powerful that they've given me the confidence to tackle more complex projects.

A few days ago, GPT-4o's image-generation capabilities took another leap forward, and the keyword "Ghibli" suddenly went viral. Many people were using ChatGPT to transform their photos into Studio Ghibli-style artwork, with amazing results!

As a freelancer, I immediately wondered if I could build a website to replicate this feature — specifically, a Ghibli-style filter tool that could convert user-uploaded images into Ghibli-style art.

Unfortunately, GPT-4o hasn't released its image-to-image API publicly, so I had to look for alternatives. I checked Hugging Face and OpenRouter, but neither offered a Ghibli-style image-to-image model.

My last hope was Replicate — and to my surprise, they had not just one but multiple models available. After testing several, I chose one with good results and a reasonable price. The model runs on Replicate's H100 hardware, costing about $0.013 per generated image.

Originally, I wanted to offer a free trial feature, allowing new users to generate 1-3 images for free. But I couldn't find a suitable model (forget about those extremely low-quality, cheap ones), so I had to scrap the trial idea. I know this might deter some curious users, but from a business perspective, I didn't want to start off losing money. Instead, I designed three credit-based pricing plans — users can buy credits and use 1 credit per generated image.

The Tech Stack I Used

• Frontend + Backend: Next.js v15
• Database: Supabase
• Image Storage: Cloudflare R2

I purchased a Next.js SaaS template, which took me a few days to get familiar with. But considering the time it'll save me in future projects, I think it was worth it. By the way, I bought my domain from Spaceship.

At first, I tried deploying the site on Cloudflare Pages but failed — Pages only runs on Edge Runtime, which imposes some limitations and prevents certain npm libraries from working. Eventually, I deployed it on Vercel, a platform I'm more familiar with. By connecting my GitHub repo and tweaking a few settings, the site went live smoothly. Vercel has a free tier, but if traffic grows, upgrading to a paid plan might be necessary.

What I Learned

1. How to use a Next.js SaaS template to quickly build a full-stack website
2. How to manage a database with Supabase (they have a free tier!)
3. How to handle image storage with Cloudflare R2 (also has a free tier)
4. How to implement Google OAuth for user logins
5. How to track website analytics with Google Analytics
6. How to add internationalization (multi-language support)
7. How to handle basic SEO
8. How to integrate Stripe SDK for payment processing
9. How to use Cursor IDE's AI-assisted coding (including Project Rules for better AI context)

Regrets

1. Couldn't find an affordable image-to-image API, so no free trial feature
2. The site has too few pages and relatively simple functionality
3. Failed to deploy on Cloudflare Pages

Next Steps

1. Add more pages (e.g., a text-to-image tool, blog posts)
2. Optimize SEO content, build backlinks, and monitor Google Search Console for organic traffic growth
3. Keep searching for better (and reasonably priced) image-to-image models to make the core feature more competitive

Thanks for reading! If you're curious, check out my site: GhibStyle. Feel free to ask any questions in the comments — I'll do my best to answer them. Cheers!