A folder of product shots pulled from three suppliers, a phone camera roll, and a legacy DAM export looks like one job from the outside. The biller sees something else entirely. Cutout work is priced per image at a size band, and the moment you mix resolutions into a single set, the invoice stops tracking the file count and starts tracking pixels. For teams running bulk jobs, that gap between “how many images did I send” and “how many units did I burn” is where budgets quietly go sideways. The good news is that the arithmetic is simple enough to run on a spreadsheet before you upload anything.
The Unit Is a Megapixel Slice, Not a File
BiRefNet, the background removal model on Y-PAD, is listed at $0.0005 per image up to 1 MP, pro rata above that, up to 4 MP — provisional beta pricing as of 7 Nov 2026, so confirm it on the live pricing page before you commit a monthly number. That single sentence contains three rules pulling in different directions: a floor at 1 MP, a straight-line ramp to 4 MP, and a published ceiling. An 800×600 JPEG at 0.48 MP is one unit. A 1500×1500 at 2.25 MP is 2.25 units. A 2000×2000 at 4 MP is four units. Nothing above 4 MP is published on the model page, so oversized captures have to be brought inside that range before they enter a batch; inputs also cap at 25 MB and accept png, jpeg, and webp.
The trap is that unit price is flat while effective rate per megapixel is not. Divide it out: a 0.48 MP image paying $0.0005 works out to roughly $0.00104 per megapixel, while a 4 MP image paying $0.002 comes to $0.0005 per megapixel. That is a two-to-one spread created entirely by size. Small thumbnails are expensive per pixel and cheap per file; large captures are the reverse. Which means the composition of a mixed set matters more than its average, and an average megapixel figure quoted in a planning doc will mislead you.
The floor is the part that catches people off guard. Y-PAD’s pricing page states that every request gets batch pricing with no minimum charge — one file costs the same per unit as a thousand — but that is a statement about requests, not about files. The 1 MP threshold is a per-file unit minimum, in the same family as Whisper’s 30-second per-file minimum or Qwen3-TTS’s 500-character minimum. Shrinking a 0.9 MP photo to 0.3 MP to “save money” saves nothing: it still bills as one unit. Only files that are already above 1 MP have anything to give back.
This is also why the pricing page’s worked example of 1,000 cutouts for $0.50 should be read as a floor, not a forecast. It describes a set where every image sits at or under 1 MP. A random mixed set rarely does.
Running the Arithmetic on a Mixed Set
Take a realistic 1,000-photo set: 600 shots at 800×600 (0.48 MP) from an older catalog, 300 at 1500×1500 (2.25 MP) from a supplier, and 100 at 2000×2000 (4 MP) from the studio. The units break down like this. The 600 small files bill one unit each, giving 600 units and $0.30. The 300 mid-size files bill 2.25 units each, giving 675 units and $0.3375. The 100 large files bill four units each, giving 400 units and $0.20. The batch totals 1,675 units and $0.8375. The headline estimate of $0.50 is off by 67.5%, and only 10% of the files caused more than half the overage.
A second shape appears when the mix is bimodal: 500 images at exactly 1 MP (500 units) and 500 at 3 MP (1,500 units) comes to 2,000 units, or $1.00. Doubling the average megapixel count doubled the bill while the file count sat still at 1,000. File count sets the baseline; pixels above 1 MP set the overage. A useful shorthand for planning is that total units land near the number of files at or under 1 MP, plus the summed megapixels of everything larger. Note that aspect ratio is irrelevant — a 4000×1000 panorama is 4 MP exactly like a 2000×2000 square. Only width × height counts.
File size in megabytes tells you almost nothing, which trips up teams that sort folders by bytes. A heavily compressed 4 MP JPEG can weigh under a megabyte; a noisy 1 MP PNG with an alpha channel can weigh three. Measure dimensions, not disk footprint. The 25 MB input cap is a separate gate that a mixed set can fail on its own terms, independent of billing.
One consolation: only successful items are charged, and failed or cancelled items cost nothing. So an oversized estimate is not the same as an oversized invoice — but it does determine how much balance gets reserved when the batch is submitted, which matters if you are running prepaid packs against a tight balance.
Shaping the Batch: Resize, Group, and Hold
The largest lever available is normalization. Take that same 1,000-photo set and resize so nothing exceeds 1 MP: every file bills one unit, the batch lands at 1,000 units and $0.50 instead of 1,675 units, and the difference is the entire gap between the headline price and your invoice. Whether that is acceptable is a quality question rather than a billing one — cut a sample of twenty photos at both sizes and look at hair, mesh, and transparent plastic before deciding. You are trading megapixels for edge fidelity, and the API will not object either way.
If the full resolution genuinely matters, group by band and submit in separate batches: one for files at or under 1 MP, one for roughly 1–2.5 MP, one for 2.5–4 MP. There are two reasons. The first is holds — at submit, Y-PAD reserves the sum of maximum units multiplied by list price, then settles at actual succeeded units and returns the unused portion. A batch of 4 MP files reserves four times the balance of a 1 MP batch with the same file count, for as long as the job runs. The second is reconciliation: a per-band batch is far easier to check line by line against a spreadsheet than a single mixed run.
Pipeline limits shape the rest. Y-PAD accepts up to 100 files per call and up to 2,000 items per batch, upload links expire after an hour, outputs stay available for seven days, and there are no webhooks — you poll. A 5,000-photo mixed set is therefore several calls regardless of how you slice it. If your pipeline chains cutout into 4× upscaling with Real-ESRGAN, remember that images carrying an alpha channel take a second pass at the same price, so upscaling a cutout costs two image-units rather than one — factor that into any size comparison you are doing.
Finally, watch the lock. The price used is stored with the batch at submission, and these are provisional beta fares that may change before launch, so an existing batch stays put while later ones may not. Confirm current per-megapixel bands against the live pricing page before sizing a quarterly budget, and check that the background removal api page still publishes the same 1 MP floor and 4 MP ceiling you built your estimate around.
Kraków-born journalist now living on a remote Scottish island with spotty Wi-Fi but endless inspiration. Renata toggles between EU policy analysis, Gaelic folklore retellings, and reviews of retro point-and-click games. She distills her own lavender gin and photographs auroras with a homemade pinhole camera.