AI image work can start with a free allowance, a monthly subscription or usage-based API billing. The price of one generated image is only part of the budget: you may also pay for rejected attempts, corrections, larger exports and a separate layout tool.
For a practical estimate, price a finished image you can use, then multiply by the number you need in the same billing period. Dreamina is a useful starting point when you want generation and reference-based editing in one image workflow; check the selected model's credit quote before each step. ChatGPT, Midjourney, Adobe Firefly and Leonardo use different allowances, so their credits, tokens and GPU hours are not interchangeable.
Prices below were checked on October 22, 2026. Each example retains its displayed currency and billing cycle. Local checkout, taxes and eligibility can change the amount you actually pay.
- The main ways you pay for AI images
- Current subscription examples—and what the allowance actually means
- A free image can consume credits and still need an export check
- Budget generation, corrections and retries together
- Calculate cost per accepted image after the period closes
- API pricing: include the reference inputs and the whole request
- A practical budget before your next batch
The main ways you pay for AI images
A credit is an accounting unit inside a particular service. A plan with 2,000 credits can be better or worse for your task than one with 1,575, depending on what the operation consumes and what is included without credits.
Likewise, an “unlimited” feature needs a model and mode attached to it. Unlimited relaxed image generation does not necessarily include fast processing, every partner model, all resolutions or API access.
Current subscription examples—and what the allowance actually means
These individual-plan examples help identify a suitable billing model. They are not a cross-currency cheapest-to-most-expensive ranking.
Use the regular renewal amount to plan a recurring workflow. Dreamina’s first-month offers are listed separately on the pricing page, so you can choose the subscription on its ongoing credits and creative benefits.
On Firefly, the premium-credit number is not a quota for all image creation: the page separately includes unlimited standard image features. Partner models and other premium operations can have different credit costs. Monthly credits do not roll over. The page's larger-plan discount advertised an October 21 end date, so those discounted prices are not used for this October 22 comparison.
Midjourney's annual Basic price works out to USD 8 per month, but the commitment is USD 96 upfront. That distinction matters if you only need a one-month campaign. Its Fast GPU allowance is also a time budget, not a guaranteed number of accepted images.
Choose the service around the work. For regular illustrated concepts, compare generation allowances. For corrections to existing photos, compare selection, reference and editing controls first; a large allowance does not repair a tool mismatch. The comparison of partial-photo editing tools can help you choose that workflow.
A free image can consume credits and still need an export check
An actual Dreamina free-account session on October 21 used Seedream 5.0 Lite, 16:9, 2K, one output, without reference images. The prompt described a coral paper boat on navy water with warm reflections and no lettering.
The Seedream 5.0 Lite operation used 6 credits, taking the balance from 28 to 22. A small generation like this is a useful way to explore a visual direction before expanding the project.
The downloaded Lite result was a 2560 × 1440 JPEG. The original file retains its visible AI and Dreamina AI markings.
For this simple illustration, six credits covered the selected Lite generation and its download. Use a similarly focused first image to explore the look before spending the rest of the allowance on variations.
Before counting a free result as finished work, open the downloaded file. Check the actual pixel dimensions, visible markings and details at the size you will use. A preview that looks suitable may not be the export your project requires.
Budget generation, corrections and retries together
A useful credit estimate is:
Credits needed = initial generation + editing attempts + rejected retries + optional upscale, summed across all the images you intend to finish.
Do this within one allowance period. Do not divide one month's subscription price by several months of output while ignoring the additional subscription payments.
For example, imagine you need 30 finished images in one month. Your selected workflow quotes 14 credits for the first generation and 14 for one correction. Those are planning assumptions based on quoted 14-credit operations seen in the existing reference-image work; they are not an audited debit for every stage.
Against the example Basic allowance of 1,575 monthly credits, the first two plans fit arithmetically; the third does not. This is a capacity estimate, not a forecast that 30 accepted images will result. It also assumes every operation really has that quote and that the allowance is available for this work.
The practical next step is to price the remaining operations before starting the full batch. If each of those 30 images also needs a separate upscale costing U credits, add 30 × U. Read U from the chosen upscaler and settings; do not assume that generating at “2K,” editing and upscaling have identical costs.
For a character sequence, keep the approved reference and prop image close at hand. When a later scene changes the prop, revise that detail before making the next image; this keeps the batch focused on one design. The character consistency guide shows how to connect those steps.
Calculate cost per accepted image after the period closes
Use two separate numbers:
Cash spent in the period tells you what left your budget: subscriptions, top-ups, API charges and any additional software required for the deliverable.
Allocated cost per accepted image helps compare workflows: divide the cost assigned to the project by the number of accepted images from that same period. State whether you allocate the entire subscription or only a project share.
For a simple example, if you assign an entire USD 20 monthly subscription to one project and accept 40 images that month, the allocation is USD 0.50 per accepted image. If you accept 10, it is USD 2.00. This arithmetic is not a promise that a USD 20 plan supplies either quantity or that its marginal price is 50 cents.
The same distinction applies to a credit subscription. An unused balance has no automatic value as finished images, and an image you reject still consumed any credits actually debited. Record both successful and rejected attempts rather than counting only the final downloads.
If your subscription also serves writing, video or other projects, use a clear allocation rule. You can report the full cash charge and the image project's assigned share side by side. Human time is separate again: add it when you need the complete production cost rather than only the tool bill.
API pricing: include the reference inputs and the whole request
API billing suits applications and repeatable production, but an output-only price is not a complete image-editing bill. ChatGPT Plus does not include OpenAI API usage; the subscription and API are billed separately.
OpenAI's current API pricing table lists these standard token rates for both GPT Image 2.5 Sunburst and Flare:
The two models can consume different token quantities at the same quality setting, so matching token rates do not guarantee matching per-image cost. Editing with reference images adds input usage; requested size and quality affect output usage. Through the Responses API, include the mainline model's token usage as well, as explained in the image-generation cost guide.
For a planning illustration, 1,000 image-input tokens, 2,000 image-output tokens and 200 text-input tokens would cost USD 0.069 at those rates: USD 0.008 + USD 0.060 + USD 0.001. Those token counts are invented inputs to demonstrate the calculation, not a measured request or a size-specific quote. Additional mainline usage, partial images, retries and applicable taxes are outside that illustration.
Cached rates are separate. For these image models, cached input pricing applies to the image tool in the Responses API, not direct Images API requests. Use the actual billing record and request usage to reconcile spending; do not assume every repeated reference receives a cache discount.
A practical budget before your next batch
Choose one representative image that needs the same kind of generation, correction and delivery as the rest of the job. Keep a short record of its selected model, settings, quoted cost, actual debit where visible, retries and downloaded dimensions.
Then estimate the whole period from that workflow. Leave enough allowance for corrections, check which operations qualify for any unlimited benefit, and confirm whether you need a separate upscale or layout step. Keep the invoice currency intact when comparing plans; converting prices without matching billing terms and output quality creates a false sense of precision.
For a small reference-based project, open Dreamina, confirm that the selected image model is available to your account and review its quote. Start the full batch once you know what an acceptable exported image—not just a first preview—requires.