Build-in-Public: What I Learned Building an AI Image SaaS

I launched AdLoft AI six months ago. It's an AI tool that turns product photos into ad creatives for e-commerce sellers. No design skills needed—just upload a photo, pick a vibe, and get 20 variations in seconds.

I built it in public on Twitter and Indie Hackers. Shared weekly updates: MRR hits, bug fixes, user feedback. This is what I learned, raw and unfiltered. If you're a solo founder eyeing AI SaaS, read this before you start.

Lesson 1: Pick a Problem You Can Solve in 48 Hours

My first version was garbage. I spent weeks on fancy features like video generation and A/B testing integrations. Users ignored them.

Then I stripped it down. MVP: one image upload, five style templates (clean, vibrant, minimalist, etc.), instant downloads. Built the core with Replicate's API for Stable Diffusion and a simple Next.js frontend. Deployed to Vercel. Total time: two days.

First 10 users signed up that week. They wanted speed, not bells and whistles. Moral: Ship a tool that does one thing insanely well. AI image gen is commoditized—differentiate on e-comm specifics, like auto-background removal for products.

Here's the stack I settled on:

Tool	Why I Chose It	Cost/Mo
Replicate	Dead-simple API, handles GPU scaling	$0.01/image
Vercel	Free tier for starters, scales easy	$20 at 1k users
Stripe	Plug-and-play billing	2.9% + 30¢
Uploadthing	File uploads without headaches	Free

Total burn at launch: $50/month. Now at $300 with 500 users.

Lesson 2: Charge from Day Zero (and Ignore the 'Freemium Trap')

Everyone says freemium for AI tools. I tried it. Got 1,000 signups, 2% converted. Churn was brutal—users grabbed free credits and ghosted.

Switched to paid-only: $19/mo for unlimited generations. Signups dropped 80%, but revenue jumped 5x. Paying users give real feedback and stick around.

Pro tip: Tier it smart.

• $9/mo: 100 images
• $29/mo: Unlimited + custom templates
• $99/mo: API access for agencies

Lifetime value now averages $250/user. Acquisition cost via Twitter threads: ~$2.

Lesson 3: Users Don't Care About Your Tech, Only Results

I geeked out on fine-tuning models with e-comm datasets. Shared benchmarks: "My model beats Midjourney by 15% on product realism!"

Crickets. Users care about their ads performing better on Facebook/ TikTok.

Shifted focus: Integrated a simple analytics dashboard. Upload your ad, track click-through rates from past campaigns. AI suggests tweaks based on winners (e.g., "Bright backgrounds convert 2x—try this").

Feedback loop:

1. User uploads photo + old ad performance data
2. AI generates variants optimized for high-CTR patterns
3. User tests, feeds back results

This turned one-off users into monthlies. Churn dropped to 8%.

Lesson 4: Building in Public Pays Rent (Literally)

Posted every milestone:

• Week 1: MVP live, 0 users
• Week 4: $100 MRR, fixed upload bug
• Month 3: $1k MRR, added mobile app

Twitter got me 40% of users. Threads on "AI ad hacks" drove 200 signups. Indie Hackers comments led to partnerships with Shopify apps.

But it's not free PR. Time sink: 2 hours/week writing updates. Trolls exist—"Your tool sucks compared to Canva." Ignore them; amplify fans.

Metrics from public building:

![Growth chart]
Alt: MRR growth from $0 to $3k in 6 months

Lesson 5: AI Costs Will Eat You Alive—Budget Ruthlessly

Image gen is cheap per use, but scales sneaky. At 10k generations/mo, Replicate bill hit $400.

Optimizations I made:

• Cache common styles (saves 60% compute)
• Queue jobs, process in batches
• Offer "low-res preview" to cut full-gen calls

Switched to self-hosted ComfyUI on RunPod for heavy users. Dropped costs 40%. Always model for 10x growth.

The Mistakes That Almost Killed It

1. Overbuilt frontend: Figma obsession delayed launch by a month. Use Tailwind + shadcn—done in days.
2. Chased virality: Referral program flopped. Focus on retention first.
3. Ignored SEO: All traffic from social. Now ranking for "AI product ad generator" after 10 targeted posts.

What's Next (and What I'd Do Differently)

Hitting $5k MRR soon. Roadmap:

• White-label for agencies
• TikTok video ads (biggest ask)
• EU data compliance for Shopify stores

If starting over:

• Launch on Product Hunt day 1
• Partner with e-comm influencers earlier
• Build mobile-first (80% users on phone)

Building in public stripped away BS. No hype, just numbers. AdLoft isn't a unicorn—it's $3k/mo helping 500 sellers make better ads. That's enough.

If you're building AI SaaS, share your wins/fails below. DM for the full tech teardown.

Current stats: $3.2k MRR, 520 users, 12% MoM growth.

---

I'm [Your Name], founder of AdLoft AI. Building tools for e-comm creators. Follow for weekly AI marketing breakdowns.