Editorial guide · September 2026

What one generated image really costs

On Promptchan, one image costs 15 Gems. That is the same number whichever pack you bought. What is not the same is the money: 15 Gems is $1.50 if you topped up with the small pack and $0.45 if you bought the large one. Same button, same day, same picture — three and a bit times the price.

Nobody is hiding this. The pack grid is public and the button says 15. But the platform prices the currency, and you price the image, and those two sums are never printed on the same screen. So the number that would actually let you decide is the one nobody has written down.

We wrote it down. Nineteen platforms, every published per-image cost multiplied by every published pack rate. On nine we could complete the arithmetic. On the other ten the sum stops, and it stops in two different places.

See how each platform scores on pricing and generated media → Independent editorial scores from the Atlas desk

An image does not have a price. It has a range.

Two numbers make the cost of a generated image, and only one of them is displayed. The first is the action cost: 15 Gems, 3 Coins, 50 XOT, one selfie credit. It sits on the Generate button, it is usually honest, and it is the number people quote to each other.

The second is what that unit cost you, which depends entirely on the top-up you happened to make — we measured that ladder across the whole market and found gaps of more than three to one inside a single platform. Multiply the two and you get the only figure that compares with anything: the cash price of one picture.

Because the second number is a range, the answer is a range too. That is not a rhetorical point. It is the difference between an image that costs less than a coffee and one that costs more than three.

What one image costs on the nine we could price

These are the platforms where both halves of the sum were published: a per-image cost we could read on the generator, and a pack price we could read on the top-up screen. The cash figures are ours — action cost multiplied by pack rate — not a number any platform displays.

PlatformOne image costsDearestCheapestWhat moves it
GoLove AI1 Star€0.30€0.093.3× — and the ladder is not monotonic, so the dearest rung is not the one you would guess
Promptchan15 Gems$1.50$0.453.3× — the platform also says the selected model changes the button price
Dream Companion3 Coins$0.30$0.122.5× — six packs, from $9.99 to $149.99
DONDIfrom 5 tokens$0.20$0.082.5× — and five tokens is a documented floor, not a ceiling
Xotic AI50 XOT$0.45$0.281.6× — three rungs, all published on a public rate card
Kindroid1 selfie credit≈ $0.14≈ $0.101.4× — but the spread here is the store, not the pack
GPTGirlfriend12 Coins, two outputs€0.36€0.271.3× on the action we ran — an earlier avatar action on the same account cost 6 Coins
DarLink AI2 tokens$0.20$0.161.25× — the top-up and the image are both in tokens; it is the plan’s Coin allowance that does not convert
OurDream AI10 DreamCoins, two slots≈ €0.12 the runa single checkout add-on, €11.99 per 1,000 — no ladder, so no gap

Per-image costs read on each platform’s own generator between 13 and 21 August 2026; pack prices read on each top-up screen over the same window, in the currency each one charged us, and not converted between currencies. Every cash figure in this table is our arithmetic — action cost multiplied by pack rate — and none of them is published by the platform. No pack was purchased anywhere, so tax, refundability and expiry are not established. Rows are ordered by the size of the gap. Terms change regularly: always check the current generator and top-up screen before you buy.

Read the last column before the first. The absolute prices across rows are worth less than they look — a Gem and a Star buy pictures at different resolutions from different models, so a cheaper image can be a worse one. The gap inside a single row is the comparable thing, and on four of the nine it is 2.5× or more.

This is ordinary volume pricing, and there is nothing improper about it. It becomes a problem because of when you buy: the small pack is what you reach for when a balance empties mid-session, and the small pack is where an image costs the most.

The pack is the biggest lever, and two platforms triple the price

Promptchan‘s grid runs from 50 Gems for $4.99 to 10,000 for $299.99. At 15 Gems an image, that is $1.50 at one end and $0.45 at the other. Nothing on the generator indicates which of the two you are about to spend.

GoLove AI reaches the same 3.3× by an odder route. Its interface documents one Star per image, which is the cleanest action cost in the corpus — but its eight-rung Star ladder does not fall as it climbs. The 500-Star pack is the best rate on the page at €0.10 a Star; the 600-Star pack directly above it is €0.13. Spending more can raise the price of every image you then generate, and there is no badge saying so.

On both, the fix is the same and it takes a minute: divide each pack price by its units before choosing a rung, not after.

Three other things move the same picture’s price

The pack is the largest lever but not the only one. Three more turned up in the corpus, and each changes the cash cost of an image you would describe in exactly the same words.

Where you buy. Kindroid publishes both its web and its app-store packs on the same page: $5.99 for 50 credits or $9.99 for 100 on the web, $6.99 for 50 or $11.49 for 100 through a store. One image is one credit either way, so the same picture costs about $0.10 or about $0.14 depending on which surface you opened your wallet in. Kindroid is the only platform we rank that shows you both prices. It is certainly not the only one where the two differ.

What you ask for. Secrets.ai‘s image studio prices resolution openly: 50 Moments at 1.5K, 100 Moments at 2K, with 4K marked as coming soon. A sharper version of the same prompt is exactly twice the money. Kindroid meters the subject instead — one credit for one person, 1.5 for two, two for three, and four to animate an avatar. GoLove AI‘s generator exposes quantity and resolution controls above its one-Star baseline, and its own pages say other media charges can be dynamic.

Which screen you are standing on. On Kupid AI, the current image Studio showed a four-coin Generate button while a separate public legacy image page still displayed “Surprise Me” at two coins. Those are differently named actions on different surfaces, so we will not claim they are the same purchase — but a reader who found the second page first would reasonably conclude an image costs half what it does.

You are not always buying one usable image

Every price above is the price of pressing the button. It is not the price of getting what you asked for, and the gap between those two is where a cheap image quietly becomes an expensive one.

On OurDream AI, a 10-DreamCoin run is presented as two images. Ours returned one safe visible result and one slot that stayed blurred or locked. If that is the normal outcome rather than our bad luck, the real rate is the whole run for one usable picture — about €0.12, not about €0.06. OurDream does refund cleanly elsewhere: two failed creator actions in the same test came back marked Refunded, which is better behaviour than most of this market documents at all.

Prompt fidelity does the same arithmetic more slowly. Kupid AI‘s one generated image was usable but missed several requested details. DarLink AI delivered in about forty seconds with weak control over clothing. Neither is a scandal, and one result establishes nothing about either platform’s reliability. Both mean the same thing for a budget: a picture you have to generate twice costs twice as much, and no pack rate on any page accounts for that.

Which is why we treat “what happens to a failed generation” as a pricing question rather than a support question. Across the corpus it is among the least documented things there is, and our protocol marks it untested wherever we have not seen a debit reversed with our own eyes.

Five platforms name the number and never price it

On five more, the generator tells you exactly what an image costs in the platform’s own currency, and the platform never tells you what that currency costs. The action is transparent; the money is not.

PlatformOne image costsWhy the sum stops there
Kupid AI4 coinsThe public pages did not disclose coin-pack sizes or prices at all
Lovescape9 ChipsChips are sold to active subscribers only, so the rate becomes visible after you have paid
Xtease9 ChipsSame currency and same operator as Lovescape — and we do not import one product’s rates into another
Secrets.ai50 Moments at 1.5K · 100 at 2KThe clearest pre-purchase disclosure in the corpus, and a top-up screen we did not open
Get Harder5 tokensA concrete button on a clear interface; standalone pack economics we did not reach

Action costs read on each platform’s own generator between 13 and 21 August 2026, before any output was produced. Xtease’s figure is our division of a displayed 18-Chip, two-image default, as is Lovescape’s. Two of these gaps are ours, not theirs: on Secrets.ai and Get Harder we did not open the top-up screen, and it is marked untested in both reviews rather than filled in.

The distinction in that note matters more than the table. Kupid and Lovescape have made the conversion impossible for anyone — one by not publishing pack prices, the other by selling the pack only after the decision to subscribe has already been taken. Secrets.ai and Get Harder simply have not been finished by us, and Secrets.ai in particular does more than most: it prints the resolution price and the video estimate before you commit, which is the opposite of the problem this guide is about.

Five do not name a number at all

On the last five, the generator did not show us what an image costs, or showed it in a currency the plan does not grant.

Candy AI lists exact image, video, voice and Live Action token costs among the things we could not establish before execution. FLIRTcam.AI confirms a Premium claim of 1,000 monthly tokens covering images, video and voice calls, but no per-image amount appeared in the state we captured. Swipey AI showed a scene control in August that we did not activate, and its current image charge was still not established on the refresh. Dreamz.ai marks images unavailable on the free tier and keeps its per-action costs in a payload rather than on a visible tariff page.

Joi AI fails differently, and it is the failure worth ending on. Its plan cards do count images: up to 20 a month on one tier, up to 75 on another, up to 350 on the top one. But the plan grants Coins while the interface spends Neurons, with no first-party conversion between the two — one of several reasons the company behind a companion can be hard to pin down. A card promising 350 images in a currency nobody can price is not a price. It is a quantity with no denominator.

★ The one-minute division

Open the top-up screen before the plan screen. Divide each pack price by its units and keep the worst result, because that is the rate you will be paying when a balance empties mid-session. Multiply it by the number on the Generate button. That is your real cost per image, and on four of the nine platforms we could price, it is more than twice what the best rung would have cost you.

Then add your own failure rate. One reroll doubles the figure, and no page on any of these sites will tell you what yours will be.

If the top-up screen does not exist, or the plan grants a currency the generator does not spend, stop there: you have not found an expensive product, you have found one that cannot be compared with the product next to it.

What a generator should show, and where to read it instead

The action cost, before you press. Fourteen of the nineteen do this, which makes it the best-observed disclosure in the category. Secrets.ai goes furthest by pricing each resolution separately, and Xotic AI publishes a rate card you can read without an account.

The pack rate, before you subscribe. This is where it breaks, and two platforms in this corpus break it for opposite reasons. A currency you can only price after paying is a price you cannot use.

One currency, not two. Where the plan grants one unit and the generator spends another, everything above stops working — for you and for us.

What a failure costs. The least documented question in the category, and the one that decides whether the number on the button is a price or an opening bid. The full per-action picture across chat, voice and video is in our guide to what an AI companion costs, the unit prices underneath it are in what a credit is actually worth, and the subscription totals they attach to are in the gap between the advertised price and the billed one.

Partly because it genuinely varies: one balance covers actions with very different underlying costs, and a credit price can be tuned without republishing a tariff page. Partly because spending 15 Gems does not feel like spending $1.50. Both are true at once. What separates a reasonable implementation from an unreasonable one is not the currency but whether its exchange rate is published before you pay — Xotic AI prints a public rate card, while two platforms in our corpus make the conversion impossible for anyone outside the company.
Usually, but not always. GoLove AI’s 500-Star pack works out at €0.10 a Star while the 600-Star pack above it works out at €0.13 — so buying the larger one makes every image you then generate about a third more expensive. Nothing on the page flags it. The ladders are short on every platform we checked, so dividing each rung yourself takes about a minute, and it is the only way to know which rung you are standing on.
Not on this table alone. We did not run a controlled quality benchmark across these nine, so a cheaper image is not an established equivalent of a dearer one — resolution, model and prompt adherence all differ, and Secrets.ai charges double for the same prompt at a higher resolution precisely because the output is not the same. What the table compares reliably is each platform against itself. Treat the cross-platform figures as a budgeting input, not a ranking.
Mostly undocumented, which is why we treat it as a pricing question. The one positive result we recorded was on OurDream AI, where two failed creator actions were visibly marked Refunded — and on that same platform a 10-DreamCoin two-image run returned one usable result and one blurred or locked slot, which doubles the effective price of the picture we could keep. Assume a reroll costs full price unless a published page says otherwise, and budget your own failure rate on top of every figure in this guide.
Because a cash price needs two published numbers, and they have one or none. Five display a per-image cost we could not convert: Kupid AI and Lovescape because the pack rate is not reachable before paying, Secrets.ai and Get Harder because we did not open the top-up screen, Xtease because we will not import a sister product’s rates. Five more never displayed a per-image cost to us at all — Candy AI, FLIRTcam.AI, Swipey AI, Dreamz.ai, and Joi AI, whose plan card counts images in a currency its generator does not spend. An incomplete table is acceptable; an invented cell is not.
They were read between 13 and 21 August 2026, which is why that window is printed under both tables. Two things move underneath them, at different speeds: pack ladders change overnight because a new rung needs no announcement, and button costs change when a model does. The structural findings last longer than the figures — a ladder that is not monotonic, a plan that grants one currency and charges in another, and a two-image run that returns one usable image are not the sort of thing that gets fixed in a week.

⚠ This guide compares published action costs and pack prices, and contains no affiliate links. The reviews it links to do: if you click through and subscribe, AI Companion Atlas may earn a commission, at no extra cost to you. Every cash figure here is our own arithmetic from prices read between 13 and 21 August 2026 — generator costs and pack ladders both change without notice, so check the current screens before you buy.