Every image API ranked for your volume
Same number of images, cheapest first.
| Model / tier | $/image | Your $/mo |
|---|
Resolution and quality drive the price
Unlike text models that bill per token, image APIs bill per generated image, and the price is set mostly by two dials: resolution and quality tier. A standard 1024×1024 render is cheap; bump to HD, 2K, or a "high quality" reasoning pass and the cost can triple or quadruple. DALL·E 3 standard sits around four cents, its HD tier near eight, while GPT-image-1 at high quality climbs higher still. At the other end, hosted open models like Flux Schnell or SDXL on Replicate can land under a tenth of a cent per image. For a product generating thousands of thumbnails a day, that spread is the difference between a hobby bill and a serious line item.
Match the model to the job
The cheapest path is almost never one model for everything. Use a fast open model (Flux Schnell, SDXL Turbo) for drafts, previews and bulk thumbnails where perfection doesn't matter, and reserve a premium model (DALL·E 3 HD, GPT-image-1, Imagen) for hero images and customer-facing finals. Add caching so you never pay twice for the same prompt, and you've cut the bill without touching quality where it counts. The table above makes the tradeoff concrete for your own volume.
When self-hosting wins
Above a few hundred thousand images a month, renting a GPU and running an open model yourself often beats per-image API pricing — the fixed GPU cost amortises across huge volume. Below that, the API's zero-ops convenience wins easily. Find your crossover on the self-hosted vs API calculator, and if images are one piece of a bigger AI product, total it all on the AI app cost estimator.
Related tools & guides
AI app cost estimator · Self-hosted vs API calculator · LLM token cost calculator · Cheapest LLM API · All AI APIs