AI Credits: What One Buys and How to Convert It

AI credits are the units AI tools use to bill each image, clip or edit. A credit is not a price until you divide it. Take the plan price and divide it by the credits included. That gives you a cost per credit. Multiply it by what the operation costs, and you have a number you can compare.

The catch is that the operation cost moves too. In DesignerBox the image or video model you pick changes the cost of each run. Almost every tool in this category quotes its price in credits. No two credits are the same size, and no two tools mean the same thing by “an image”.

This covers the conversion, why there is no single cost per image, what raw generation costs at list, what a video clip costs, and the three terms that quietly move your bill.

Key Takeaways

• Convert before you compare. Plan price divided by credits gives a cost per credit. Multiply by the credits that operation costs. Now two quotes are comparable.
• There is no single cost per image. The model sets it, so always name the model when you quote a cost per image.
• A credit is a fraction of a plan. Two tools can quote the same credit number for one image and still charge different money for it.
• The video model moves the price most. Test on a lower-cost model, and spend the premium rate only on the shots you keep.
• Raw model list prices are public and low. Seedream 5.0 Lite lists $0.035 an image, FLUX1.1 [pro] $0.04, and Nano Banana 2 $0.067 at 1K, on the vendors’ own API pricing pages (September 2026).
• Three terms move your bill quietly: retries charged as generations, credits that reset each month, and a model switch that sits behind one dropdown.

How do you convert AI credits into a real price?

Turning AI credits into a real price takes one division and one multiplication.

1. Cost per credit = plan price divided by the credits included.
2. Cost per operation = cost per credit multiplied by the credits that operation costs.

The example uses two made-up tools, because the method is what you need.

Tool B charges fewer credits for the image and twice the money. The credit number on its own tells you nothing. A headline promising the lowest credit count can be a smaller number attached to a bigger bill.

Three things break the division if you skip them. Read the billing period, because a page that opens on the annual view shows a lower per-month figure than the monthly plan charges. Read whether the allocation resets or accumulates. And read whether purchased packs expire.

In DesignerBox every run shows its cost before you start it, so you rarely do the division above by hand. Plans and credits are on the pricing page. If you are pricing the whole job and not one image, start from the four stages and what each one runs on.

Why there is no single cost per image

The model does the work, and the models cost different amounts. In DesignerBox the credit price belongs to the model you pick, and the word “image” has no price of its own. The image models are on the model list, and each one has its own cost.

Some operations have a fixed cost. An avatar run returns nine fixed poses for 25 credits.

This has two consequences for a budget. Any plan you build has to name the model, because “500 images” costs different amounts on different models. And the default model matters more than the most expensive one. Nano Banana Pro runs unless someone picks another model, so plan around its cost.

What does raw generation cost at list?

Published list prices from the model vendors’ own API pricing pages. Google’s rows were checked on 7 October 2026 and the other rows in September 2026:

Two things follow. Images are cheap at the model layer and have been for a while, so an image price is mostly not about compute. And the spread inside video is the whole story. Veo 3.1 with audio lists close to ten times Kling 2.5 Turbo’s silent 720p rate per second, and that sits behind a dropdown a junior can change. OpenAI bills its image model by token, and GPT Image 2 API pricing converts that to a cost per image. What a low cost per variant means for ad results is in AI generated ads: what 369 million impressions show.

The Veo rows have a date on them. Google lists 22 October 2026 as the earliest shutdown date for the Veo 3.1 preview models in the Gemini API, and names Gemini Omni Flash as the replacement (Gemini API deprecations, October 2026). The Nano Banana 2 row has a note too. Google released Nano Banana 2.1 on 6 October 2026 and deprecated Nano Banana 2 the same day. It told developers to move to the new model and announced no shutdown date for the old one (Gemini API changelog, October 2026). Check those rows again before you budget on them.

This is a floor. Treat it as a reference point. Run raw APIs and you also run the retries, the storage, the brand rules, the review pass and the plumbing between models. A platform price above list is buying that layer. Knowing the floor is how you judge whether it is buying anything. Our guide to why AI video credits run out covers the retry side.

What does a second of video cost?

The cost of a second of video depends on the model. At the model layer, the vendor list prices above run from $0.042 to $0.40 a second for the video models listed. In DesignerBox an eight-second clip costs 40 to 560 credits, depending on the model. To put money against it, divide your plan price by the credits your plan includes. Then multiply by the credits the run shows.

That is a wide spread for the same eight seconds. It is the largest cost decision in video, and one choice decides it.

Read it as buying power. The same monthly credits buy many more clips on a lite model than on a premium one.

The working rule has three steps. Plan the shots on a storyboard first. Test the sequence on a lower-cost model. Then spend the premium rate only on the shots you keep. What AI video generation costs explains video pricing in more detail, from credits to the wider market rates.

What three terms quietly change your bill?

Retries charged as generations. This term matters most, and it is easy to miss on a pricing page. If a generation you reject is billed, your real cost per usable asset is the list price multiplied by your attempts per keeper. At three attempts per keeper, a $0.067 Nano Banana 2 image at list is a $0.20 image, and a $0.134 Nano Banana Pro image is a $0.40 image. Ask directly whether rejected outputs are charged.

Credits that reset. A monthly allocation that resets means uneven usage wastes the allowance. A team that produces a campaign in week one and nothing after it pays for a month and uses a week. DesignerBox credit packs are a separate one-time purchase. You can buy them with an active paid plan, and they expire after 12 months. Packs work differently from the monthly allocation, so keep the two apart.

A model switch behind one dropdown. The video spread and the image spread both sit in one model choice. Anyone on the team can change the cost of the next hundred runs. Set the default on purpose, and write down which model a recurring job runs on. In DesignerBox you pick the model for each step when you build the workflow, and every run after that uses it. A template comes with its model already picked.

None of these terms is hidden on purpose. They are missing from the headline number, and the headline number is what people compare.

A fourth term sits outside the arithmetic: what the license lets you do with the result. Some free tiers permit personal use only, as the free AI video generator plans show. If the work goes into a paid ad, the cost per credit on such a tier does not matter. We read the license terms tool by tool in our guide to DaVinci AI alternatives.

What should you compare between tools?

Compare AI tools on six questions, and only the first two are about the sticker price.

1. What is the cost per image, converted, and on which model? A cost per image with no model named is not an answer.
2. What does a clip cost, by model? Ask for the cost on each model, because an average hides the spread. Video models differ in cost by more than ten times.
3. Are rejected generations charged? This sets your real cost per usable asset.
4. Do credits reset, and do purchased packs expire?
5. What is gated above your tier, and how many seats does the plan carry? A cheap credit on a plan that cannot run the job is not cheap. DesignerBox has three gates. Uploading your own photos and the commercial license start on the Pro plan. AI video, virtual try-on, upscaling, the image editor and the video editor start on the Premium plan. Team features, shared brand kits and white label are on the Ultra plan, and every plan below Ultra is one seat.
6. What does the price buy beyond generation? In DesignerBox: Assets, the file library, the saved workflows, the brand profile the workflow reads on every run, and the critic steps that score results before your own review. A price above raw list has to cover this layer.

A credit price you cannot divide is not a price. Our guide to the DesignerBox plans explains what each plan includes. If a tool charges for rejected results, the reject rate decides your real cost.

The cost of a run is one half of the answer. The other half is how much of the job the price covers. The full workflow from the first product photo to the finished ad, in one subscription. In DesignerBox that means the templates, the workflows, the apps, batch, the image editor, the video editor, your brand rules and the models each step runs on. See how the four stages fit together.

FAQ

How much does one AI-generated image cost?

At the model layer, the vendors’ own list prices in September 2026 include $0.018 an image for Seedream 5.0 Flash and $0.035 for Seedream 5.0 Lite (BytePlus). FLUX1.1 [pro] lists $0.04 (Black Forest Labs), and Nano Banana 2 lists $0.067 at 1K (Gemini API). On a credit platform, divide the plan price by the credits included and multiply by the credits the run costs. In DesignerBox the cost of each run is shown before the run, so you do not need to calculate it.

How do I compare credit prices between AI tools?

Convert both to a cost per operation, and name the model on each side. Plan price divided by included credits gives a cost per credit. Then multiply by what one image or one clip costs in credits on the model you would use. Credit counts are not comparable between tools, because no two credits are the same size and no two tools mean the same thing by one image.

Why does AI video cost more per output than an image?

Video cost scales with length, and the video models are far apart in price. In DesignerBox the same eight-second clip ranges from 40 to 560 credits across the models. Picking the model is the budget decision. AI video starts on the Premium plan.

Do AI tools charge for failed generations?

Some do, and it is the most important term to check. If rejected outputs are billed, your real cost per usable asset is the list price multiplied by your attempts per keeper. On demanding work that can be three attempts or more for each keeper. Ask the vendor directly. Why an AI video generation fails lists what three model vendors say about billing for a failed run.

Do AI credits roll over month to month?

It depends on the platform, and often on whether they are subscription credits or purchased packs. Monthly allocations commonly reset, so uneven usage wastes part of the allowance. DesignerBox credit packs are one-time purchases. You buy them with an active paid plan, and they expire after 12 months. They sit separately from the monthly allocation, which resets.

What is the cheapest way to generate AI images at volume?

Raw model APIs are the floor, from $0.018 to $0.134 an image for the models listed above, and you then own the retries, storage, brand rules and review. On a platform, the lowest-cost route is a lower-cost model for the first look. Save Nano Banana Pro for the finished frame. The right comparison is cost per usable asset, including retries.

How many credits do I need per month?

Count the results you need, then name the model for each. Run a small test set on the model you plan to use, and note the credits each run uses. Then multiply by your monthly volume. Count video separately, because the video model moves the cost more than any other choice.

Sources

• Google, Gemini API pricing for Nano Banana 2.1, Nano Banana 2, Nano Banana Pro, Veo 3.1 and Veo 3.1 Lite, standard tier (ai.google.dev, accessed October 2026)
• Google, Gemini API deprecations, shutdown dates for the Veo 3.1 preview models (ai.google.dev, accessed October 2026)
• Google, Gemini API changelog, Nano Banana 2.1 release and the Nano Banana 2 deprecation on 6 October 2026 (ai.google.dev, accessed October 2026)
• ByteDance, BytePlus ModelArk pricing for Seedream 5.0 Flash and Seedream 5.0 Lite (docs.byteplus.com, accessed September 2026)
• Black Forest Labs, FLUX1.1 [pro] documentation (docs.bfl.ai, accessed September 2026)
• Kuaishou, Kling API video pricing (kling.ai, accessed September 2026)
• DesignerBox pricing page (designerbox.ai/pricing), September 2026: plans, credit allocations, credit packs and feature gates

Model list prices verified from each vendor’s own API pricing page as of September 2026. Google’s list prices, its shutdown dates for the Veo 3.1 preview models and the Nano Banana 2 deprecation were re-checked on 7 October 2026. Model pricing changes frequently. Individual results vary.