The subscription is the advertised number. For anyone who generates pictures regularly, it is frequently the smaller half of the bill.
None of this is a scam. It is the direct pass-through of a real cost difference, and it becomes manageable the moment you do one piece of arithmetic before subscribing rather than after.
Why the meter exists
A text reply costs the operator a fraction of a cent. An image costs meaningfully more — GPU time, measured per generation. Voice is billed per second of audio. Video is dramatically more than either.
That is the entire explanation for the pricing shape you see across the category: generous or "unlimited" messages, capped images, voice sold by the minute, video on upper tiers only. As covered in how the media side works, the chat and the pictures come from different engines with very different cost profiles.
So when a pricing page says unlimited, read it as unlimited text. It is usually telling the truth about the cheap thing.
The three ways apps charge for images
Included in the plan, capped. A daily or monthly allowance inside the subscription. Easiest to reason about: you know the worst case, which is the subscription price.
Credits, bought separately. The plan covers conversation; images draw on a balance you top up. The worst case is unbounded, which is the entire problem.
Tiered allowances. More images at each price point. Reasonable, as long as you price the tier you will actually use rather than the entry one.
The first is why Candy AI leads our ranking at $12.99 — chat, images and voice inside one subscription removes this entire decision. That is a pricing architecture advantage, not a feature one.
The arithmetic to do before you pay
Two numbers, five minutes.
One: how many images do you actually want per week? Be honest and slightly generous. People underestimate by a factor of two in the first month, when the novelty is highest.
Two: what does that cost on each model?
- On an inclusive plan: the subscription. Done.
- On credits: images per week × 4.3 × the per-image price in the bundle you would realistically buy, plus the subscription.
If the credit answer exceeds the inclusive answer — and above roughly a handful of images a week it usually does — the inclusive plan is cheaper even when its headline price is higher. This is the specific case where the more expensive subscription is the cheaper product, and it is the most common mistake in the category.
How credit bundles are priced
Always the same shape: a small bundle at a high per-unit price, a large bundle at a much lower one. The gradient is the point — it is designed to move you up the ladder, and the large bundle is genuinely better value per image.
Two things to check that are rarely on the purchase screen:
Do credits expire? Often. A large bundle bought for the per-unit discount is not a discount if half of it evaporates.
Do failed generations charge? A refused prompt, a filter block, a generation you immediately discard — in many apps all three still count. Over a month of experimenting with prompts this is a real number, and it is the single most common complaint about credit systems.
The first month distortion
Expect to generate far more images in month one than in month four. The novelty is real and it fades.
This matters for the decision in two directions:
Do not buy a large credit bundle in week one on the strength of week-one usage — you are extrapolating from the peak.
And do not buy an annual plan chosen for heavy media use on the same evidence. The trade-off is worked through in monthly, annual or credits.
Where the money actually goes
Worth knowing so the caps feel less arbitrary: you are paying for GPU time, and you are paying for it whether or not you like the result. An app that gave away unlimited image generation would be underwriting the difference from the subscription, and that business ends.
Apps doing more on the media side — Secret Desires generates images and short video, and sits at 4.3 in our ranking — carry a correspondingly higher cost base. Generosity with media is a real cost decision, and it shows up in the plan shape rather than in the marketing.
The short version
Count the images you want per week, price both models, and prefer the one where the worst case is a number you already know. Then check expiry and whether failures charge.
Five minutes of this beats any comparison table, including ours.
