I registered tryonr.com on November 24, 2025 — a Monday night, around 11pm, after my third cup of coffee. Looking back at the notes I scribbled that night, I think I was wrong about roughly half the things I believed in that moment. No team. No funding. Just a hunch about AI virtual try-on that I couldn't shake.
Five months later, AITDK's third-party estimate puts Tryonr at roughly 8.9K monthly visits. That's not a "growth-hacker case study" — it's "barely above noise" territory. But the curve is up and to the right, and the AI virtual try-on bet behind the project is more intact than ever.
This is an honest, slightly uncomfortable look at building an AI virtual try-on tool from zero. I'll share the numbers I almost didn't share, the bet I made on AI try-on, and the five things I'd tell myself the night I registered the domain.
The $2,000 Problem That Started Tryonr
A friend I'll call Huang (name changed for privacy) runs a small Etsy clothing store. Last December, Huang told me he'd just spent $2,000 on his latest product shoot. Twelve products. One rented studio. One hired model. Two weeks of post-production.
The shoot paid for itself. Sales jumped 40%. But Huang adds new products every month. That's $24,000 a year not going to inventory, not going to ads, not going to his family. For a store doing $8,000 a month in revenue, that math is brutal.
That's the moment Tryonr started. Not in a pitch deck. In a real conversation about a real cost that AI virtual try-on was finally good enough to replace.
The Bet Behind Tryonr (And the Time I Quit)
Here's something I don't usually tell people. I tried to build this in 2023. The models weren't ready. Hands were broken. Faces drifted between angles. I gave up after three weeks and shelved the idea for 18 months.
Then late 2025 happened.
For years, AI-generated try-on images of people wearing clothes looked wrong. Manual fixing cost more than hiring a photographer. But by late 2025, three model releases finally crossed the AI virtual try-on commercial threshold at the same time:
- Kling 3 — Human motion and consistency strong enough for video try-on workflows. (My full Kling 3 guide.)
- Seedance 2 — Character consistency across multiple shots, finally usable for product try-on. (How Seedance 2 compares to Sora and Kling.)
- Nano Banana 2 — Image editing that respected clothing texture and brand details.
None of these models was built for AI virtual try-on specifically. But together, they meant a small seller could finally generate model shots that looked good enough to sell from. Not perfect. Good enough.
That's the window Tryonr was built for.
What Tryonr Actually Is (And Isn't)
Tryonr is not a model. I'm not training anything from scratch — no GPUs, no data team, no $50M Series A.
What Tryonr does is wrap best-in-class AI virtual try-on models behind workflows designed for one specific user: a small e-commerce seller who needs model shots and try-on visuals without paying $2,000 per shoot. (Here's the full $2,000-to-$1 workflow.)
That's the whole product.
What I deliberately said no to:
- A general AI image editor — too crowded, no defensible angle
- A social filter app — wrong distribution model for a solo builder
- A model trained from scratch — physically impossible at my scale
Saying "no" to those was the hardest part of month one. The hardest "no" I actually said: in month two, someone offered me $500 to build a custom AI fashion mood-board feature. I needed the money. I said no anyway, because it had nothing to do with virtual try-on. That was the cheapest expensive decision I've made on this project.
The Numbers (Including the Ones I Almost Hid)
I drafted this post twice without the numbers. Both drafts felt like the kind of bullshit I hate reading from other founders. So I'm including them.
Here's what AITDK estimates for Tryonr over the first three full months of indexed content:
- February 2026 — near zero monthly visits
- March 2026 — around 2.5K monthly visits
- April 2026 — around 8.9K monthly visits
A few honest caveats no founder story usually includes:
- These are third-party estimates, not my own GA. Treat them as directional.
- Bounce rate is around 51%. People come, look, sometimes leave. The on-site experience is something I'm still working on.
- Average session is short. Pages per visit: 1.7. Not great.
- Global rank is around 2.4M. That sounds bad — and it kind of is. Most of the long tail of the web is noise, but I'm not "winning" anything yet either.
I'm sharing these numbers because I would have found a post like this useful six months ago, when I was searching for what realistic early-stage indie AI virtual try-on growth actually looks like. Most of what I read back then was either too vague ("it grew slowly") or too rosy ("we 10x'd in a month"). The truth is messier.
5 Things I'd Tell Myself in November 2025
If I could send a note back to November 24, 2025 at 11pm, this is what it would say:
1. Pick the moment, not the trend
I didn't start Tryonr because "AI is hot." I started because three specific AI virtual try-on model releases changed what was technically possible at a specific moment. Look for capability shifts, not buzz peaks. Buzz lags; capability leads.
2. Content is the only distribution I can afford
No paid acquisition. No audience. The only thing I could do was write — guides, comparisons, reviews of the very AI virtual try-on models Tryonr is built on. Most traffic comes from those posts. Write about the tools that make your tool possible.
3. Most posts won't work — and that's fine
I've published dozens of posts in five months. Maybe five drive meaningful traffic. The rest don't.
One specific example: my first attempt at a generic "best AI fashion tools" listicle picked up only a handful of visits in two months. The very next post — a focused, step-by-step Kling 3 guide — picked up hundreds of visits in a single week. Same effort. Same author. Different specificity. That's the lesson I keep relearning.
4. Don't fake the numbers. Don't hide them either.
The post you're reading almost didn't include traffic data. Then I realized: hiding numbers is just a slower way of being caught. Anyone evaluating an indie AI virtual try-on tool will figure out the scale anyway. Honesty is cheaper than the alternative.
5. Stay narrow or die wide
Every "feature creep" temptation outside virtual try-on, I tried to remember why I said no to general image editing in month one. Scope discipline isn't a productivity tip for a solo builder — it's survival.
What's Next for Tryonr
I'm not promising dates. Indie projects that promise dates miss them.
What I am doing over the next few months:
- Writing more about the AI virtual try-on models that make this whole space work
- Tightening the workflow for the seller use case (less generic, more specific)
- Adding new try-on models as they cross the quality bar
- Fixing the on-site experience so 51% bounce becomes 35%
If you're a small e-commerce seller using Tryonr — the on-site experience is what I most need feedback on right now. The 51% bounce rate tells me something is broken, and I'd rather hear it from you than guess. Reply to this post or email support@tryonr.com. I read every message.
The Bottom Line
Five months in, here's what I'm sure of: indie AI is winnable. But only by the people willing to tell the truth about how slow it actually is.
If you're starting something in AI right now: pick a thin slice, write about it, and don't lie about your numbers. That's all I've got.
Thanks for reading. The Tryonr blog is where I keep figuring this out — and if you want updates when I write the next honest post, subscribe to the newsletter below. I send roughly one email a month.
