Count the image work, not just the number of photographs
An image budget can look small when it is estimated from the number of source files. A provider may instead charge according to the pixels processed in each generated output. One photograph can therefore create several billable pieces of work.
This example uses an entirely fictional contract and prices. It is a budgeting exercise for a catalogue, not a statement about a real provider or a promise that every image pipeline charges this way.
Write down the outputs that serve a task
The catalogue owner approves two derivatives for each of 100 photographs: a 1,600 by 1,200-pixel detail image and a 400 by 300-pixel card image. The large version lets a reader inspect the documented surface detail. The small version supports browsing a collection without downloading the larger file.
The approved dimensions are output requirements for this example. Their necessity is reviewed against the actual presentation; they are not universal recommendations for every device or photograph.
The fictional provider counts every generated output's pixel area, divides by one million, and sums the result without rounding each derivative upward. It charges only the demonstrated transformations in this example, not the unchanged stored source. Reprocessing counts again. Storage and delivery are separate terms.
Calculate the measured processing unit
The detail image contains 1,920,000 pixels, or 1.92 megapixels. The card contains 120,000 pixels, or 0.12 megapixels. Each source therefore requires 2.04 processed megapixels for these two outputs.
For 100 sources, the initial approved generation is 204 megapixels. The fact that there are 100 original files does not turn this into 100 processing units. Nor does stronger file compression reduce the pixel area of an output whose width and height remain unchanged.
The organisation expects one initial generation and one approved full replacement during the budget period. Its complete forecast is consequently 408 processed megapixels. The record also states that unexpected reprocessing would add usage; it is not hidden inside the allowance.
Compare complete fictional offers
Plan A includes 300 processed megapixels for 30 budget units, then charges 0.10 per additional megapixel. The forecast overage is 108, costing 10.80, so the predicted processing charge is 40.80.
Plan B includes 500 processed megapixels for 38 budget units. The same 408 forecast fits its allowance. In this fictional comparison, the owner verifies that both plans provide the required transformations, output formats, storage arrangement, delivery terms, access and support. No missing capability or separate mandatory charge reverses the comparison.
The decision selects B for this forecast and these terms. It is not a rule that a larger plan is always cheaper. With only the first generation, A's 30 could be lower than B's 38, which is why the expected replacement work matters.
Do not make quality disappear to fit a budget
Someone suggests producing only the card image and stretching it in the detail view. That would reduce processing volume, but the controlled visual review no longer allows the intended surface detail to be examined. The owner rejects that option because it fails the approved reader task.
Another suggestion reduces the number of generated files by serving the large version in every card. That changes delivery and performance costs, which must be compared separately. A smaller transformation bill is not a complete result if it makes browsing unnecessarily expensive or slow.
The chosen scope keeps both derivatives. The owner also requires generation to occur when an approved source revision changes, rather than recreating the same outputs every time someone opens a page. The budget is based on the observed generation policy, not a claim that all caching strategies behave identically.
Reconcile the acceptance example
The fictional usage record contains 100 detail outputs and 100 card outputs for the first approved run, totalling 204 megapixels. The replacement run contributes another 204. Their combined 408 agrees with the forecast; a read-only page visit adds no transformation in the inspected setup.
An unchanged control image retains its two output hashes when no generation is requested. A deliberately repeated approved generation is recorded as additional processing rather than disappearing from the accounting. These checks distinguish forecast, actual transformations and ordinary viewing.
The remaining limit is uncertainty about future revisions and traffic. The owner assigns someone to compare usage with the forecast and review the plan before a larger campaign. The exercise does not establish real provider pricing or the future number of edits.
When discussing scope and cost, bring the output dimensions, generation policy and complete provider terms. The performance service can help connect image decisions to the reading task, so a saving is judged alongside the experience it must preserve.