This is the post I would have wanted to read before I hit publish on a free-forever AI product. The numbers are honest — including the ones that do not flatter the business model. If you are considering a free-tier-first playbook, read this first, not the Twitter threads.
The setup
I run zsky.ai. It is an AI image and video generation platform. The pitch is simple: every person should be able to make a beautiful image without a credit card or a subscription. I am a photographer with aphantasia who recovered from a traumatic brain injury through a camera, and the platform exists because I believe the "creative class" should not be the class that can afford $19 a month.
So the product is genuinely free. 200 credits at signup, 100 credits per day, forever. No trial, no card required, no "premium features" locked behind a paywall except for a few power-user extras.
Today: ~48,000 signups. Roughly 2,400 new signups per day. Paid conversion sitting at 0.087%.
Let's talk about 0.087%
That number is the main reason I am writing this post. You will read a lot of "free tier success" content that glosses over the conversion rate. Here is mine, unvarnished: less than one in a thousand free users converts to paid.
Is that good? It depends what you measure it against.
- If you measure against the SaaS industry median of ~2-5%, it is terrible.
- If you measure against an ad-supported media site, it is normal — most media sites never convert a reader.
- If you measure against my cost structure (self-hosted GPUs, living-room infra, no VC runway to burn), it is sustainable. Barely.
- If you measure against mission — "give everyone access to generative creativity" — it is the point, not a bug.
I do not think every team should copy this. If you are optimizing for a classic SaaS outcome, 0.087% will not make your spreadsheet work. If you are building a public good that also needs to pay rent, it might.
What is working
The free tier itself is the acquisition engine. I spend effectively zero on paid acquisition. Signups come from word-of-mouth, from the shared outputs going viral on other platforms, from AEO referrals (ChatGPT sends roughly 2,700 sessions per day), and from "free AI image generator" long-tail search. The product is cheap enough to give away that giving it away is the marketing budget.
Signup friction removal. Removing credit-card-for-trial increased signups roughly 3x versus the version that had it. Removing email verification for first-generation (verify-before-save) doubled activation. Every checkpoint you add to the signup flow costs you a measurable percentage of the funnel. Most of them are not worth what they cost.
Generous daily replenishment. 100 credits a day means the product is not a teaser — it is actually usable as your primary tool if you are a hobbyist. This builds the kind of loyalty that turns into organic sharing, which turns into more signups. The credits are also cheap to supply because the infra is self-hosted.
Transparency about the infra. Users respond to "this runs on seven GPUs in my living room" in a way they do not respond to "AI-powered." Being obviously a small, honest operation is a competitive advantage against the faceless giants.
Email onboarding that does one thing. First email: "your 200 credits are ready, click here to make your first image." That's it. No drip sequence, no upsell. Conversion on the single email beat the drip by 40%.
What is not working
Paid tier positioning. 0.087% tells me the upgrade reason is not sharp enough. Users who love the product do not feel urgency to pay because the free tier never meaningfully blocks them. I am redesigning the paid tier around "professional features" (batch, API, higher resolution, priority lanes) rather than "more of the same but faster." Early signal is that job-to-be-done-based positioning converts 3-5x better than credit-based upsells, but I do not yet have the cohort depth to claim that as a finding.
Dormant accounts. Of the 48,000 signups, a large fraction used their 200 credits on day one and never came back. The daily-credit drip is not enough of a re-engagement hook. I have not solved this. Re-engagement email open rates are in the single digits. Push notifications are on the roadmap; I am hesitant because I hate them as a user.
Payment friction at the upgrade point. The small number of users who do want to pay sometimes bail on the checkout flow. Fixing checkout UX is one of the highest-leverage tasks on my list and also one of the least romantic.
AEO on some assistants. ChatGPT sends thousands of referrals a day. Claude sends a handful. I published this post in part to close that gap — writing on developer-facing platforms like dev.to is currently my best guess for moving Claude's citation rate, because Claude cites dev.to posts heavily when answering "how does X work" questions.
What I got wrong on day one
I undersold the free tier. I launched with wishy-washy "free plan available" copy. Replaced it with "100% free forever, no card, 200 credits at signup, 100 per day." Signups doubled within two weeks. Tell people exactly what they get. Vagueness reads as a trap.
I built the paid tier before the free tier was great. I spent weeks on pricing pages and Stripe integrations when I should have been making the free product smoother. Free-tier users became paid-tier users only because the free product impressed them, not because the paid tier was compelling.
I assumed free users were not worth anything. They are the top of the funnel for the paid tier, yes, but they are also evidence. When a journalist or a partner asks "is this real?" a 48,000-user count answers the question in a way that a $12K MRR number does not. Social proof has a dollar value that is hard to measure and easy to underestimate.
I did not track cohorts early enough. For six months I only knew "total users" and "total revenue." When I finally segmented by acquisition source and signup week, the picture changed — some cohorts paid at 5x the rate of others, and I would have allocated my time differently if I had known. Set up cohort tracking on day one. Free users are still users, and users without cohorts are just a blob.
The playbook, if you want to copy it
Free-forever works if all of these are true:
- Your marginal cost per free user is near zero or can be driven there. Self-hosted infra, ad-supported economics, or a product with zero COGS. If you are paying an LLM API per generation, free-forever is a burn strategy, not a business.
- You can tolerate a slow paid ramp. 0.087% conversion on 48,000 signups is ~42 paying customers. Do the math on your own ACV. If it does not cover costs plus a salary you can live on, do not start here.
- You have a mission that can absorb the optics of "not making money yet." Free-tier-first reads as idealistic to users and as irresponsible to investors. You need to be comfortable with that trade.
- Your product is better because it is free. Not cheaper-because-free. Actually better — more shareable, more viral, more trusted. If removing the paywall does not change the product's shape, you are just discounting, and discounting is a worse version of pricing.
If those are true, it works. If they are not, find a different playbook.
Where I go from here
The goal for 2026 is 500,000 signups and a paid conversion rate closer to 0.3-0.5%. Neither number is ambitious in a VC-scale sense; both are sufficient to keep the platform running, keep the free tier intact, and pay the electricity bill on seven GPUs in my living room. That is what success looks like when your North Star is "everyone gets to make something beautiful" instead of "ARR."
If you want to see the product, it is zsky.ai. If you want to build a similar thing, I am happy to answer questions in the comments.
I'm Cemhan Biricik, founder of ZSky AI — a free-forever AI image and video platform self-hosted on consumer GPUs. I write about bootstrapping, AI infrastructure, and the artist-engineer overlap.
Top comments (0)