The AI Image Generator API for Every Top Model
Revana puts the image models developers ask for behind one API key and one request shape: send a prompt, name the model and get the image back in seconds. You pay per image from prepaid credit, an image that fails costs nothing, and switching models is a one-word change in your code.
# One Idempotency-Key per job, made once: resend with the same $KEY after a timeout, a 429 or a 5xx
KEY=$(uuidgen)
curl https://api.revana.si/v1/generations \
-H "Authorization: Bearer $REVANA_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $KEY" \
-d '{
"model": "nano-banana-pro",
"input": {
"prompt": "A ceramic coffee mug on a white table, soft window light, catalog photo",
"resolution": "1K",
"aspect_ratio": "4:5"
},
"max_cost_usd": 0.04
}'Pick the right image model for the job
Photoreal products and people
Nano Banana Pro is $0.04 an image at 1K and 2K, Imagen 4 $0.05. Both give lifelike product shots and portraits, and Nano Banana Pro takes up to 14 reference pictures to keep a product or a face the same across a set.
Readable text on the picture
Ideogram V3 (from $0.03 in turbo mode) and GPT Image 2 ($0.03 even at 4K) handle posters, packaging and social graphics where the words matter as much as the picture.
Drafts and big batches
GPT Image 2.6 Flare starts at $0.02 an image and Qwen Image 3.0 at $0.03. Draft on a low-cost model, then render the final picks on a premium one; each image is billed on its own.
Style, concept art and characters
Midjourney V7 and Seedream 5.0 Lite lean toward strong styles, moody scenes and consistent characters. Seedream 5.0 Lite is $0.03 an image at 2K and 3K.
Prices on Revana as of November 2026. Every model, size and price is on the models page.
One request shape for every image model
Send
Post a model name and an input object to /v1/generations: a prompt, an aspect ratio and a resolution, plus reference picture URLs where the model takes them. The field names are the same on every model, and a value a model does not support is refused before anything is charged.
Receive
Poll GET /v1/generations/{id} or receive a signed webhook when the job finishes. The finished job returns the file URLs in output.urls and the exact charge in cost_usd; nothing is hidden in a separate billing report.
Switch
To try another model, change one word in the request: your parsing, storage and billing code stay the same. That makes side-by-side tests cheap, and new models show up on Revana soon after release.
Several pictures from one request
Set num_images to get up to 4 pictures from one prompt. Each delivered picture is billed at the model's price, so four Seedream 5.5 pictures at 1K cost $0.08. Midjourney V7 already returns four pictures per request.
What 1,000 images cost
| Model | Setting | Per image | 1,000 images |
|---|---|---|---|
| Seedream 5.5 | resolution: 1K | $0.02 | $20.00 |
| Grok Imagine Image 2.0 | resolution: 1K | $0.02 | $20.00 |
| GPT Image 2.6 Flare | resolution: 1K | $0.02 | $20.00 |
| GPT Image 2 | resolution: 1K | $0.03 | $30.00 |
| Flux 2 Pro | resolution: 1MP | $0.03 | $30.00 |
| Ideogram V3 | quality: turbo | $0.03 | $30.00 |
| Qwen Image 3.0 | resolution: 1K | $0.03 | $30.00 |
| Nano Banana Pro | resolution: 1K | $0.04 | $40.00 |
| Imagen 4 | any size | $0.05 | $50.00 |
| Recraft V4.1 | resolution: 1K | $0.05 | $50.00 |
Prices as of November 2026. Bigger sizes cost more on most models: Nano Banana Pro is $0.08 at 4K and Seedream 5.5 $0.06 at 2K. The Nano Banana Pro calculator prices any batch.
Failed images cost nothing
More than one route
Most Revana models can be reached through more than one route. If the first errors, times out or runs out of capacity, the next takes over, and you write no retry code.
Charged once
The price is held when you submit and released if the job fails. You are charged once, for the image you get back, so batches stay predictable even at thousands of images a day.
Capped and idempotent
Send an Idempotency-Key so a resend never makes a second job, and max_cost_usd so no request can cost more than you planned.
Edit, upscale and animate on the same key
Edit a picture
Send an image and an instruction such as "change the background to a beach at sunset". Nano Banana Pro, Flux Kontext Pro and Qwen Image Edit handle it, with the same key and billing as generation; the image editing API guide compares them.
Upscale the keepers
Run a finished image through Recraft Crisp Upscale for $0.01 for a sharper, larger file. Upscale only the images you keep, so drafts stay cheap.
Turn a still into a clip
When a still is ready, animate it with Kling 4.0, Seedance 2.5 or Veo 3.1 on the same key; image and video jobs share one balance and one bill. The AI video cost calculator prices the clip.
A complete example in Python
Submit, poll and save two pictures to disk, with a cap on the whole request:
import os
import time
import uuid
import requests
API = "https://api.revana.si/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['REVANA_API_KEY']}"}
# One Idempotency-Key per job, made once: resend with the same key after a timeout,
# a 429 or a 5xx and you get the same job back, never a second charge.
key = str(uuid.uuid4())
resp = requests.post(f"{API}/generations", headers={**HEADERS, "Idempotency-Key": key}, json={
"model": "seedream-5.5",
"input": {
"prompt": "A running shoe on pale concrete, hard noon shadows, catalog photo.",
"resolution": "2K",
"aspect_ratio": "4:3",
"num_images": 2,
},
"max_cost_usd": 0.11, # your cap for the whole request
})
resp.raise_for_status()
gen_id = resp.json()["id"]
job = requests.get(f"{API}/generations/{gen_id}", headers=HEADERS).json()
while job["status"] not in ("succeeded", "failed"):
time.sleep(3)
job = requests.get(f"{API}/generations/{gen_id}", headers=HEADERS).json()
if job["status"] == "failed":
raise SystemExit(job.get("error")) # a failed job is not charged
for i, url in enumerate(job["output"]["urls"]):
with open(f"shoe-{i}.png", "wb") as f:
f.write(requests.get(url).content)
print("charged", job["cost_usd"])Make images from Claude, ChatGPT or Cursor
Add the Revana MCP server to Claude, ChatGPT, Cursor or VS Code, and your agent can list models, get a quote and generate images in the same chat, on your prepaid credit. Ask for five product shots on a white background: it picks a model, shows the price, waits for your yes and returns the files.
Your first image in five minutes
- Create a key. Sign up with Google or an email link and create an API key in the dashboard. The key is free and works with every model, and price quotes are free too.
- Add credit. Credit is prepaid from $10, with no subscription and no monthly minimum, and it never expires. One balance covers images, video, music and language models (pricing).
- Send the request. Copy the sample at the top of this page, add your key and name a model. The API docs cover webhooks, reference pictures and batch jobs, each with a working example.
FAQ
What is an AI image generator API?
It lets your app send a text prompt, and optionally reference pictures, to an AI model and get a finished image back. Revana gives you one API key for Nano Banana, GPT Image, Seedream, FLUX, Midjourney, Imagen and other image models, with one request shape for all of them.
How much does an image cost?
It depends on the model and the size. On Revana Seedream 5.5 starts at $0.02 an image and Nano Banana Pro costs $0.04 at 1K (prices as of November 2026). Every model's price is on its page, and failed images cost $0.
Is there a free way to try the API?
Your Revana API key is free, and so are price quotes, so you can test requests and check costs without paying. Images are paid per image from prepaid credit, which starts at $10, never expires and has no subscription.
Which model is best for text in images?
Ideogram V3 and GPT Image are the strongest choices for readable text such as posters, labels and social graphics. Both run on your Revana key, so you can test the same prompt on each and keep the result that reads best.
Can I get several images from one request?
Yes. Set num_images to get up to 4 pictures from one prompt on models that support it; each delivered picture is billed at the model's price, so four Seedream 5.5 pictures at 1K cost $0.08. Midjourney V7 returns four pictures per request.
Why use one API instead of each model maker?
Calling each maker directly means a separate account, key, request format and bill for every model. With Revana you integrate once, and if a route fails the job moves to another route for that model, so a failed image costs nothing and needs no retry code.
How fast are images?
Most images come back in a few seconds; larger sizes and premium models can take a little longer. Jobs run asynchronously, so you can poll for the result or receive a signed webhook when each image is ready.
Can I use the images commercially?
Commercial use depends on each model maker's terms, so check the terms for the model you plan to use before publishing. Each model page on Revana names its maker.
Start generating images today
Create your free Revana key and send your first prompt in minutes. Pay only for images you get back, and ask the Revana team about volume pricing: a real person replies.