Looking for a Leonardo AI Alternative? Tokens vs Coins, Compared Before You Pay

You are on the pricing page of one tool with another open in a second tab, and the question is not which one makes nicer pictures. It is which billing model fits how you actually work, and whether the thing you want to make is allowed. This post answers both for Leonardo and for us, in that order, and names where they win.

Everything here about Leonardo was read on 3 October 2026. Their main website refused our requests that day, so we could not read their plan prices and have not guessed them. What follows comes from their public help center and their API documentation, and each claim links the page it came from. If their pages say something else by the time you read this, their pages are right.

What Leonardo offers, as checked on 3 October 2026

Leonardo runs on tokens. A free account gets 150 tokens a day, reset every 24 hours and lost if unused. Paid plans, named Essential, Premium and Ultimate, carry a monthly token allowance; unused tokens go into a rollover bankthat holds up to three months' worth, with a cap per plan, and only while the subscription is active. Top-up tokens can be bought on top and do not expire, but again only with an active subscription. Premium adds an unlimited slower mode for images once tokens run out; Ultimate extends it to video. There is no free trial of the paid plans.

On their token cost table a default 768x768 image is one token, an upscale five, their premium pipeline eight to sixteen, and each reference image adds a few more. The toolset around that is wide: an editing canvas, a realtime canvas, training your own style models on your own images, upscalers, background removal, video with start and end frames, 3D generation and mobile apps. Their API is a separate product billed pay-as-you-go in US dollars, with a balance that does not expire.

Two lines in the commercial-use article matter more than the table. Commercial use is allowed on every plan, free included. But private generation is a paid feature: on a free account everything you make is public, visible to and usable by other members, and grants Leonardo a perpetual licence to it. Deleting images also requires a paid plan. On content, their FAQ describes prompt and output filters, says free users are filtered more strictly, and asks users to avoid prompts that lead to pornography; their API blocks such prompts by default.

What Leonardo does better

The editing. Leonardo is a workbench, not just a prompt box: you can paint over part of an image, extend it, remove a background, upscale it in two flavours, and do all of that inside one canvas without leaving the tab. We generate a picture and hand it to you; the nearest thing we have to an editor is a reference image. If your work is iterating on one image rather than producing many, that difference outweighs everything below.

Then the trained models: upload your own pictures and get a style or a subject you can call by name in later prompts. We offer references per generation, not a model that remembers them. The free daily allowance lets you learn the tool without paying, the higher plans' unlimited mode makes the marginal image free for a heavy user, there are mobile apps, and a very large public feed to borrow prompts from. The per-token price on a plan you spend to the end is lower than ours per image, and we would rather say so here than have you discover it afterwards.

What Fellowi does differently

• No subscription. You buy coins, and a generation in the image studio costs a fixed number of them set by quality and options. Coins never reset, there is no month to use them up in, and there is no plan to keep active for the ones you already bought (an image generator without a subscription).
• Failures refunded, automatically. A generation that fails on our side returns its coins without a ticket, and a prompt refused before it runs is never charged (what a failed generation costs you). Their refunds page says refunds are assessed manually by a support agent.
• Commercial use with no credit and no plan. Our terms claim no ownership of what you generate and place no restriction on commercial use: sell it, print it, use it in client work. We take no share and require no attribution (the commercial-use clause).
• Private by default, for everyone. Nothing you generate is shown to other users and there is no paid tier to switch that on.
• Character generation for adults. For users of legal age our studio allows it, with hard refusals for anything involving minors or non-consent. Their moderation is built to block it.
• Reference images as a normal line on the same price sheet, your own photo included (editing with your own photo).
• Video with sound from the same coin balance, with optional start and end frames, on the video page.
• A public API on the same wallet, same price, same refunds, no separate account to fund.
• Eleven languages. The studio, the help and this blog are written in each; their FAQ says the interface and support are English only.

The cost, like for like

Compare the unit, not the headline. Their unit is a token, and a default image is one; a plan carries thousands, so on a month you spend to the end the price per picture is very low, and lower than ours. Ours is a coin, and the coins per image at each quality, with each option, are on the product page, generated from the code that charges; we do not retype them into posts because they would be wrong the day a price moved. What changes the sum is the month you do not spend: their tokens roll over for up to three months and only while you keep paying, ours sit there for as long as you like. Count the images you made last month, look at both numbers, and the answer is arithmetic. The same exercise against a flat plan is in the Midjourney comparison.

Who should pick which

Take Leonardo if you edit as much as you generate, if you want a model trained on your own pictures, or if you make images every day and will spend the plan. Take Fellowi if you generate in bursts and resent paying for the quiet months, if you want private output and commercial rights without a subscription, if failed generations should refund themselves, or if what you want to make is a character for an adult audience. Many people will be right to use both: one to produce, one to retouch.

Where we fall short

We have no editing canvas: no painting over a region, no extending a frame, no background removal. We do not train models; a reference image guides one generation and is forgotten by the next. Generated files are kept for a limited retention window and then deleted, so download what you want to keep rather than treating your history as an archive. And we do not warrant that you hold copyright in what you generate; in several countries a purely machine-made work may not qualify, and that is the law, not our policy (who owns AI-generated images). If any of those is the deciding factor, the other tab is the right one. If none is, open the studio, make one picture, and decide with it in front of you.