As checked 2026-10-25: Some services charge per generated image, others consume internal image output tokens, credits, GPU time or subscription fast-mode minutes. This is a dated economics guide, not a permanent price guarantee or a personal hands-on benchmark.
Scope: global English reference; USD list prices where shown; local eligibility, currency, taxes, negotiated terms and model release dates require separate confirmation. Dynamic source-page prices have a check date, not a verified initial effective date.
Why the price per image is elusive
Some services charge per generated image, others consume internal image output tokens, credits, GPU time or subscription fast-mode minutes. Even where a vendor publishes an equivalent per-image rate, that number may refer to a particular resolution or default quality. Google’s Gemini 3 Pro Image page distinguishes 1K/2K from 4K output equivalents; fal lists charges by model endpoint and output unit; Midjourney describes included fast GPU time under each subscription. These are different denominators and cannot be averaged into a meaningful universal price.
Use finished deliverables as the denominator
For a campaign producing 200 accepted product illustrations, define accepted as meeting resolution, brand, legibility and rights requirements. Record how many outputs, upscales, edits and human revisions were needed. If 320 images are generated to obtain 200 accepted assets, the unit cost per accepted image uses total generation charges divided by 200, not divided by the 320 raw outputs. Those numbers are merely a scenario illustration, not a claim about any named model’s success rate.
Break down settings and service tiers
Maintain separate records for text prompt inputs, reference images, output resolution, aspect ratio, quality setting, seed or variation, upscale, batch route and commercial license plan. Distinguish vendor API charges from a creative subscription’s monthly price. An application that invokes an external model through a third-party routing service may pay the router’s schedule rather than the model developer’s direct API rate.
Avoid the obvious mistakes
Do not assume all retries are unbilled, that a subscription grants unlimited fast generations, or that an advertised free trial includes commercial rights. Unlimited relax-mode access can involve queue or concurrency rules and is not equivalent to a guaranteed number of output images per hour. If the cost of an uncommon setting is unpublished, request a live quote or leave it unpriced.
Practical selection
For occasional creators, a subscription may simplify spending; for automated volume pipelines, metered per-output rates and instrumentation are more auditable. Price is only one axis: record editing features, reliability, desired resolution and the time spent obtaining a usable result. The imaging team’s technical evaluation should remain separate from this financial model.
A multi-resolution campaign
A publisher might need a small article thumbnail, a social preview and a high-resolution printable image from the same concept. Even where a vendor accepts one prompt for all three, its resolution tier and edit/revision path can create separate billable operations. Model the three deliverables explicitly and record the dimension and image output SKU. A price per default 1K image says little about a final 4K poster. Keep the number of outputs accepted after legibility and rights review separate from the number delivered by the API.
Quality gates that affect money
Establish rejection criteria before generation: unreadable text, inaccurate object design, failed reference consistency, unwanted logos, unsuitable face likeness, wrong aspect ratio or resolution. Compare two workflows by mean billable generations per accepted output, including upscales and edits, while acknowledging human-review variation. Do not claim that one generator ‘always follows text better’ without a repeatable evaluation. Paid privacy features and license restrictions can outweigh a superficially cheaper output unit, especially for client-confidential material.
Decision checklist
- Identify the specific plan or API endpoint, model version, region and currency before pricing a job.
- Measure input, output, caching, generated media and failed-job units separately.
- Check the official pricing page again before procurement, especially if the SKU has no effective-date notice.
- Keep source evidence and a timestamped price ID; do not treat unknowns as zero.
Pricing evidence boundary
This cost discussion concerns Multi-vendor. Reproduce a quote only with its exact model or plan identifier, chargeable unit, billing region, processing mode and timestamp; a changing vendor rate card is not a permanent price guarantee. If any one of these dimensions is missing, keep the associated rate and worked scenario under editorial review instead of filling the gap with an estimate.
Sources & Further Reading
- Gemini Developer API shows tier/model/context/modality-specific rates and cache storage — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- Midjourney named tiers, subscriptions, fast GPU time, relax and stealth conditions — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- fal model-route output unit and GPU-hour list tariffs — official first-party source, accessed 2026-10-25; original effective date not specified in source.