100 product images bill as calls, including the ones you discard

Flat-billed image models charge a fixed price per completed call — aspect ratio, resolution, and prompt length are not terms. 100 keepers is 100 calls only if every prompt lands; a 50% keep rate doubles the leftover /cost/100-ai-product-images meter. Per-megapixel rows are a different formula.

The formula is credits = price_per_image × images_generated. images_generated is every call that completes, including the ones you throw away. There is no volume tier and no refund for a cull. Keep rate is the term that actually decides the bill.

The meter is completed calls

Seventy-four image models bill one fixed price per generation. The cheapest option on those rows runs from 1 credit to 30 credits. Eighteen of them sit at 4 credits, the figure the cost page uses when the batch is unspecified. Multiply that sticker by the number of calls.

What does a batch of 100 AI product images cost? is the worked book: 100 images at 1 credit is 100 credits; at 4 credits is 400; at 30 credits is 3,000 — before retries. A Monthly plan’s 580 credits is 145 images at the typical 4. Call it one 100-shot catalogue set at a realistic keep rate.

Text-to-image is the generate surface. Generate an image from text is the named agent job: prompt in, picture out, priced per model before you confirm. Neither door refunds a discarded take. The charge attaches to the generation.

A 1:1 thumbnail and a 16:9 hero from the same flat-billed model cost the same. So do a first attempt and a fourth. Test the prompt on a handful of SKUs, then run the batch. At 1 credit a call the testing is nearly free. At 30 it is worth doing deliberately.

Z Image Turbo is a 1-credit draft row. GPT Image 1.5 is 2 credits, 4K, and takes text-to-image, image-to-image, and edits — each as its own call. Pick the row. Count the calls.

Nine things that do not change the price

On a flat-billed image model the sticker is the whole price. These knobs are composition choices, not billing terms:

1. Aspect ratio. Square, story, landscape — same call price.
2. Resolution. A 1024px output and a 4K output from the same flat row cost the same. Size is free here.
3. Prompt length. A 400-word brief with three references costs what a five-word one costs. Prompt tokens are not this meter.
4. Reference images. Not metered on these rows. They still have to be your product.
5. Render time. Slow or fast, the completed call is one unit.
6. Whether you keep the file. The cull does not refund the call.
7. Whether you delete the file. The generation already billed.
8. Batch volume. 500 images cost five times 100. No bulk discount.
9. Edit versus generate. An edit model is its own catalogue entry, billed per call. It is not a surcharge on a previous generation.

That list is why “we only needed 100 keepers” is not a budget. Write the keep rate down before you dispatch.

A 50% keep rate doubles the leftover meter

Keep rate is how many generations survive the cull. Pick a pessimistic one for a first shoot. The cost page’s retry table:

• 100% keep — 100 calls for 100 keepers
• 75% keep — 134 calls
• 50% keep — 200 calls
• 33% keep — 304 calls
• 25% keep — 400 calls

A 50% keep rate doubles the budget without changing a single price. At the typical 4 credits, 100 keepers are 400 credits if every prompt lands and 800 credits if half of them do not. The worked example on the cost page is a 100-shot set at 4 credits with a third of the shots redone: 4 × 133 = 532 credits.

Do not average a 4-credit row with a 30-credit row and call it “about 10.” Model choice is a fixed multiplier on every image. The spread is 30× — 200 credits versus 6,000 on 200 calls. Prompt discipline is the only cost control this family has.

The cost hub is the rest of the map.

Per-megapixel rows are a different formula

Not every still is a call. Thirty-three image models bill per megapixel of canvas: credits = rate_per_megapixel × (width × height ÷ 1,000,000). Doubling both edges quadruples the charge. A discarded print-scale render still costs the area price. That leftover is print-resolution AI image cost, and the square-law writeup is doubling both edges quadruples a per-megapixel bill. Do not mix the pages.

A handful of image rows use resolution_flat: one price up to HD, a higher price at 4K. That is a two-step band, not this family.

What a per-call charge does not cover, even on a flat row:

• Upscaling a chosen image afterwards, billed per the upscale model you select.
• Background removal and colour work, each billed as its own image generation.
• Turning stills into video, which moves you onto a per-second meter entirely.

Those are extra calls. Count them separately. Lock the prompt on five SKUs at the cheap end, then run the hundred. A 50% keep on a 4-credit row is still cheaper than a 100% keep on a 30-credit row.

FAQ

Does a 4K product image cost more than a 1024px one on these models?

Not on a flat-billed model — one price per call whatever the output dimensions. Models where size drives price are the per-megapixel and resolution-banded families. Those have their own pages.

Am I charged for images I delete?

Yes. The charge attaches to the generation, not to what you do with the result. That is why keep rate rather than sticker dominates a batch budget.

Is there a bulk discount for 100 or 500 images?

No. The formula is a straight multiplication with no volume tier. 500 images cost five times what 100 cost on the same model, including the ones you discard.

Should I use a per-megapixel row for a 100-image catalogue?

Only if you need print-scale area and you accept the square law. For listing photos, a flat-billed row plus a keep-rate budget is the leftover meter this page is for. Compare both formulas before you mix them.