Image generation APIs you can pay per call

What image generation APIs do

An image generation API turns a text prompt into a picture. The caller sends the prompt and, depending on the route, a size or aspect ratio, a model, a style or a number of images. The response is an image as bytes, a base64 string or a hosted URL. An agent can use it to illustrate an article, draft a product mockup, make a thumbnail or produce an icon without a designer in the loop.

The word covers several products. Text-to-image starts from words. Image-to-image editing starts from a picture you supply and changes it. Background removal and upscaling transform a result after it is generated. A route that renders a QR code or returns a list of style keywords produces an image or a menu, but it is not generating a picture from a prompt, so it stays out of this comparison.

The model behind the route shapes the result as much as the price does. Providers name model families such as FLUX, Nano Banana, GPT Image, Krea, SDXL, Seedream or Ideogram. A model name is the provider's statement, and two services naming the same family can still run different versions, settings or safety filters. Judge them on sample images from your own prompts.

Delivery matters to an agent. A synchronous route returns the image in the same response. An asynchronous route returns a job id and a status URL, and the image arrives later. A hosted URL can expire, so a workflow that needs the image later should download and store it. Rights and content rules come from the provider and the model; check them before using an image commercially.

Compare the returned output, scope and limitations

The text-to-image type includes every route that generates a picture from a prompt, and it is the only type that can qualify a service and set its row price. Editing, transparent cutouts, upscaling, asynchronous jobs and side-by-side model runs are listed as separate types because they change what you receive or how you receive it. A route can belong to several types: a premium tier can be text-to-image, transparent, upscaled and asynchronous at once.

Video clips, songs, prompt cheat sheets and style lists can sit on the same host as an image route. They are different products with their own prices, so they cannot supply an image quote. A gateway that sells chat, speech and images in one place is priced here on its image route, not on its cheapest text call.

Model names come from endpoint descriptions. CoinopAI ImageGen, CoinOpAI Kronos Market Intelligence, ThumbnailCreator and GPUOps AI Inference name a FLUX model. ThumbnailCreator name Nano Banana. ThumbnailCreator, GPUOps AI Inference and Grawwww Studio — AI Character Video & Image API name Krea, SDXL, Seedream or Ideogram. 3 describe a choice of aspect ratio on a matched route. These are provider statements about what runs behind the route, not a test of image quality.

Each price is the cheapest endpoint of that service for that kind of data: the same number wherever this page quotes it.

Compare 9 image generation services

Each row keeps the selected text-to-image endpoint and its reference price. The order follows observed paying wallets for the service; it does not rank image quality.

$/CALL is the price of the endpoint named under it: the cheapest image generated from a text prompt endpoint of the service or, for a service without one, its cheapest image generation endpoint. Never the cheapest tool of a gateway. Other kinds of data have their own price in the tables below. IMAGE GENERATION ENDPOINTS counts the image generation endpoints out of all the endpoints the service lists. BUYERS 30D counts distinct wallets that paid the whole service on-chain in 30 days, not only its image generation endpoints. "Generalist" marks a gateway that sells more than three unrelated kinds of tools: it is never called the cheapest.

How to choose an image generation endpoint

Write the output you need

Name the size or aspect ratio, the file format, transparency and the number of images before comparing routes. A square social image, a 16:9 thumbnail and a transparent product cutout are different requests, and a route that cannot return one of them is not cheaper for that job.

Test with your own prompts

Keep a small set of prompts that cover your real use: a product, a person, text inside the image, a style you depend on. Run them on each candidate and compare the results side by side. A model name alone does not tell you how a route handles hands, lettering or brand colors.

Read the price unit

A route can price per call, per image or per image and size. A side-by-side route returns one image per model in one call. Divide the quote by the images you keep, and ask the route for the quote of your exact request when the description says the price depends on size or count.

Plan for delivery

Check whether the image comes back in the response or through a job and a status URL. For an asynchronous route, budget for polling and for a job that fails after payment. Download hosted images you need to keep, since a provider's link can expire.

Test the operating limits

The monitored median uptime is 100% and the median selected quote is $0.05. Monitoring latency describes our checks, not generation time, which can be far longer for a large or upscaled image. Measure the time to a finished image on your own prompts.

Choose by the output your workflow needs

Start with the result your next step can use. The candidates below use the same type classification as the comparison table. Each quote belongs to the lowest priced endpoint of the requested types, with the path kept.

Generate an image from a prompt

Fix the size or aspect ratio, the number of images and the output format before comparing quotes. Keep the prompt, the model named in the response and the returned URL or bytes together, because hosted links can expire.

Candidates today, most buyers first: Cluster Protocol ($0.02/call, generalist, 719 buyers) · MapleAI ($0.001/call, 81 buyers) · CoinopAI ImageGen ($0.25/call, 15 buyers) · CoinOpAI Kronos Market Intelligence ($0.25/call, 13 buyers) · AmanChain Relay ($0.007476/call, generalist, 3 buyers) · ThumbnailCreator ($0.26/call, 2 buyers) · GPUOps AI Inference ($0.02/call) · Grawwww Studio — AI Character Video & Image API ($0.05/call) · Compacthost AI Tools ($0.05/call)

Edit an image you already have

Check how the route accepts the source image and what it is allowed to change. Store the input image reference with the result so the edit can be reviewed.

Candidates today, most buyers first: MapleAI ($0.001/call, 81 buyers)

Create a transparent asset

A cutout replaces a separate background-removal step. Check the edges on a sample of your own subjects, because hair, glass and shadows are where removal fails.

Candidates today, most buyers first: CoinopAI ImageGen ($0.35/call, 15 buyers)

Get a print or banner-size image

An upscale adds pixels to a generated image; it does not add detail the model never drew. Compare the final resolution and the turnaround of the job, not only the price.

Candidates today, most buyers first: CoinopAI ImageGen ($0.5/call, 15 buyers)

Compare several models on one prompt

A side-by-side call buys one result per model. Count the images it returns and keep the model id with each one before choosing a model for repeated use.

Candidates today, most buyers first: ThumbnailCreator ($0.26/call, 2 buyers)

What the providers describe

These paragraphs attribute output contracts and limits to each provider. Reference prices follow the matching endpoint type rather than an amount written in a marketing description.

CoinopAI ImageGen

CoinopAI ImageGen splits its product into tiers. /generate describes FLUX Schnell text-to-image with a choice of aspect ratio and a hosted image URL, and its text-to-image quote starts at $0.25. /generate/clean adds background removal into a transparent PNG at $0.35. The HD and Pro tiers start asynchronous jobs: the paid response returns a job id and a free status URL to poll, and the upscaled quote starts at $0.5. The provider also sells a static prompt-style sheet on the same host; that is a text document, not an image, and it does not set the generation price.

4 image generation endpoints of 5 · $0.25/call (GET /generate) · Base · uptime 30d 100% · 221 ms · 15 buyers in 30 days

ThumbnailCreator

ThumbnailCreator describes one call that generates YouTube thumbnails with several image models side by side, naming GPT Image, Nano Banana, FLUX, Seedream, Ideogram and Grok. Its side-by-side quote starts at $0.26. The endpoint lets the caller choose the models, one to four images per model and a format such as 16:9 or 9:16, so one paid call can return several images. The provider says the call is charged only when every image succeeds; that is its stated policy, so check the result of your own calls before relying on it.

1 image generation endpoint of 1 · $0.26/call (POST /api/x402/generations) · Base · uptime 30d 100% · 1029 ms · 2 buyers in 30 days

MapleAI

MapleAI is a reseller of model access that lists /api/v1/images/generations and /api/v1/images/image2image next to chat routes. Both describe pricing per image and size, so the listed text-to-image quote of $0.001 is a reference floor rather than the price of every request: a larger size or more images can cost more. Ask the route for the quote of your exact request before you budget a batch. The editing route at $0.001 takes an image you supply, which makes it a different job from generating one from a prompt.

2 image generation endpoints of 8 · $0.001/call (POST /api/v1/images/generations) · Base/POL/SOL · uptime 30d 99.9% · 195 ms · 81 buyers in 30 days

Grawwww Studio — AI Character Video & Image API

Grawwww Studio describes /api/render/image as a single Krea 2 image rendered on a local RTX 4090. The call is asynchronous: it returns a job id and a free status URL, and the PNG is fetched from an asset URL when the job finishes. Its text-to-image quote starts at $0.05. A video route on the same host renders clips and is not part of this image quote. Plan for polling and for a job that fails after payment, and keep the job id with your record.

1 image generation endpoint of 2 · $0.05/call (GET /api/render/image) · Base · uptime 30d 99.1% · 521 ms · 0 buyers in 30 days

Cost per image and for a larger batch

The median selected route costs $0.05 per call. At the same request shape, 100,000 calls total $5,000.00. Use the generated table as a reference calculation, then count how many calls one accepted image really takes.

Each service at the price of its main image generation endpoint, the one named in the table above.

Image work rarely ends with one call. A prompt is revised, a result is rejected, a second size is needed or a cutout follows the first draft. Count the attempts behind each image you keep, and keep the exact endpoint, parameters and quote with every attempt, so a cheap draft route is not mistaken for the cost of a finished asset.

Set a budget per image and a stopping rule for retries. Separate a refused prompt, a failed job and an unavailable upstream in the result record. The provider's terms decide whether those outcomes are charged. Read the payment challenge for the exact request before authorizing it, and do not estimate image costs from a gateway's cheapest chat or embedding call.

Inspect a request and its payment evidence

For text-to-image, CoinopAI ImageGen documents GET /generate, with a reference quote of $0.25. Read its current schema to pass the prompt and aspect ratio you need, and verify that the response returns the hosted image URL and fields your workflow stores.

The listed method and description establish the advertised contract. This explanation does not reproduce an HTTP 402 challenge or certify a paid response. Open the service page's payment evidence and obtain the challenge for the exact request before authorizing payment. Compare resource, method, input, network, asset and amount together; do not reuse a challenge from a neighboring tier or style route.

How this comparison is built

The comparison starts with services that are online and payment-ready, have a known price, a description of at least 80 characters and measured 30-day uptime of at least 90%. Task selection then reads the documented endpoints. Image words in a service description can make a service a candidate without proving that a captured route generates a picture from a prompt. Pages below the index threshold can still be read directly.

Provider claims were reviewed against the public service and endpoint descriptions captured on October 22, 2026. They remain provider claims. Monitoring supplies operating figures and reference prices; it does not judge image quality, safety filters or usage rights. Buyer counts describe observed paying wallets for the service, with the coverage limits explained on its linked page.

The endpoint classification supplies the comparison quote, per-type prices and job candidates. A row needs a fixed quote for a text-to-image route. Editing, cutouts, upscales, asynchronous jobs and side-by-side runs are described as separate types, and video, music, QR codes and style lists stay out of the quote. Keep the named path when comparing services: two routes on the same host can return very different products.

Questions

How to pay

Each service answers an unpaid call with HTTP 402 and its price; your agent signs a USDC payment on the listed network and retries. No API key, no account, no subscription. See what x402 is.