Levels order the models by difficulty and required skill. They are not revenue promises.
11 models
Coding and selling are «or less» filters: picking «some» keeps the models that need none. Several of these models need no coding at all.
These orderings answer «can I start this», not «how much does it pay». We do not rank models by revenue — there is no observed distribution to rank by.
「Hours, sold in slices」
Attach AI to work you can already do, raise throughput, and sell it per job. The least to learn and the fastest route to cash.
No verified cases yet — no revenue distribution shown.
「Charging what the photoshoot cost」
Turn one product photo into scene and model shots, and sell that. What AI removes here is the photo shoot, so the buyer compares you against a studio quote rather than another freelancer.
No verified cases yet — no revenue distribution shown.
「The reader is not the payer」
Publish on a site you own and earn from ads and affiliates. There is no unit price here — revenue is views × RPM, not one sale for one job. The risk points the opposite way from the models above: nothing until it works, then it keeps arriving while you sleep.
No verified cases yet — no revenue distribution shown.
「Nothing until you clear the bar」
Post short videos and get paid by a platform programme. The three platforms ask for completely different things — YouTube wants scale, TikTok wants length, Instagram wants relationships. “Post lots of short videos and ad money arrives” is not true on any of them.
No verified cases yet — no revenue distribution shown.
「Selling the setup, not the AI」
Build chatbots fitted to a client's own documents. What is sold is not AI but the configuration against one company's material — that is the whole model.
No verified cases yet — no revenue distribution shown.
「Same date every month」
Turn one-off delivery into a monthly engagement. This is where the same work starts producing repeat income.
No verified cases yet — no revenue distribution shown.
「Riding on someone else's revenue」
Take over the listing, product-page and keyword work a seller repeats every week. In a 2026-08-20 count on Kmong, this term carried the largest total review volume of everything measured, though at a low list price.
No verified cases yet — no revenue distribution shown.
「Build once, sell repeatedly」
Ship small-business websites from a single reusable template. In a 2026-08-20 count on the Korean marketplace Kmong, this category had the highest median review count of everything measured, at a list price well above the median.
No verified cases yet — no revenue distribution shown.
「Taking the repetition nobody wants」
Find a client's repetitive work, automate it, and charge for the build plus upkeep. What you sell is not the tool but knowing the work.
No verified cases yet — no revenue distribution shown.
「Credentials set the price」
Select and implement a model for a client's problem. The highest price band on this list, and the highest barrier — what sells is a research and engineering track record, not tool operation.
No verified cases yet — no revenue distribution shown.
「Build once, charge many times」
Sell small software that solves a narrow problem, on subscription. The longest route and the highest failure rate — and the only band where revenue arrives in a month you did not work.
No verified cases yet — no revenue distribution shown.
This is not a ranking of earnings. Not one listed model has a verified revenue figure. It ranks the ceiling you cannot pass alone, computed from published prices, published delivery times and official platform fees.
We divide 20 working days by the published delivery time, then assume the month is filled with whichever tier leaves the most. It does not mean the orders arrive.
The highest-priced model is not first. The most expensive tier is ₩1,900,000 in “AI automation agency”, yet “AI product imagery service” tops the list. Turnover sets the ceiling, not price — how many you finish in a month matters more than what one costs.
Not “low” — “unknown”. Putting a model we cannot compute at the bottom would invent information, so we leave it out entirely.
⚠️ This is not a recommendation order. Being first does not make a model easy to enter, and being last does not make it bad — the lower entries often demand more credentials. Operating profit here is also before tools, your own labour and income tax.