Use GPT Image 2, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst with one API key. ¥0.25 (about $0.037) per image, the same at 1K, 2K and 4K, up to 14 reference images, failed generations not charged.
Three GPT Image models sit behind the same endpoint as the Nano Banana models. Pick one with the model field.
| Model | API id | Good for | Price per image (1K / 2K / 4K) |
|---|---|---|---|
| GPT Image 2.5 Flare | gpt-image-2.5-flare | Fastest 2.5 model, about 45 s. A good default. | ¥0.25 ≈ $0.037 |
| GPT Image 2.5 Sunburst | gpt-image-2.5-sunburst | Highest detail, about 100 s. | ¥0.25 ≈ $0.037 |
| GPT Image 2 | gpt-image-2 | Previous generation, many aspect ratios. | ¥0.25 ≈ $0.037 |
VELIN has its own small REST API (a multipart POST /api/generate, then GET /api/task/{id} to poll). Pick the model id, the aspect ratio (size, e.g. 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3) and the resolution (1K, 2K, 4K). There is no separate quality setting.
VELIN charges ¥0.25 (about $0.037) per image whatever the resolution. OpenAI bills GPT Image by tokens: GPT Image 2.5 (Sunburst and Flare) is $30 per million image output tokens, GPT Image 2 is $15 per million (OpenAI API pricing, read 2026-09-29), plus text and image input tokens. The cost per image therefore depends on size and quality. OpenAI's own image guide lists these GPT Image 2 output costs per image (image generation guide, "Calculating costs", read 2026-09-29):
| Size and quality | OpenAI GPT Image 2 (output only) | VELIN |
|---|---|---|
| 1024×1024, low | $0.006 | $0.037 |
| 1024×1024, medium | $0.053 | $0.037 |
| 1024×1024, high | $0.211 | $0.037 |
| 1024×1536 or 1536×1024, low | $0.005 | $0.037 |
| 1024×1536 or 1536×1024, medium | $0.041 | $0.037 |
| 1024×1536 or 1536×1024, high | $0.165 | $0.037 |
Fair reading: OpenAI's low-quality tier is cheaper than VELIN's flat price, its high-quality tier is several times more, and larger sizes use more tokens. VELIN is a good fit when you want one predictable number for finished images at 2K or 4K, or want a per-image invoice with no token maths. Input tokens for your prompt and reference images are extra on OpenAI's side. Prices change; check the linked pages. The USD figure for VELIN assumes about ¥6.7 per USD.
Volume: capacity is up to 100,000 images per day; 10,000 images cost ¥2,500 (≈ $373) at the flat price. Volume pricing on request. Nano Banana comparison: Nano Banana API pricing.
curl -X POST https://72agi.com/api/generate \
-H "Authorization: Bearer $VELIN_API_KEY" \
-F "prompt=a quiet room, cream light" \
-F "model=gpt-image-2.5-flare" \
-F "size=16:9" \
-F "resolution=1K"
# -> 202 {"id": "02a38b1d1b66", "status": "queued"}
curl https://72agi.com/api/task/02a38b1d1b66 -H "Authorization: Bearer $VELIN_API_KEY"
# -> {"id": "02a38b1d1b66", "status": "processing"}
# -> {"id": "02a38b1d1b66", "status": "succeeded", "url": "/outputs/02a38b1d1b66.png", "price": 0.25}
curl -o image.png "https://72agi.com/outputs/02a38b1d1b66.png"
Node 18+ (no dependencies):
// Node 18+ (built-in fetch). export VELIN_API_KEY=your_key
import { writeFile } from "node:fs/promises";
const BASE = "https://72agi.com";
const auth = { Authorization: `Bearer ${process.env.VELIN_API_KEY}` };
const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
// 1) submit (multipart form; add form.append("refs", blob, "name.png") for up to 14 reference images)
const form = new FormData();
form.set("prompt", "a red fox in fresh snow, golden hour");
form.set("model", "gpt-image-2.5-flare");
form.set("size", "16:9");
form.set("resolution", "1K");
const submit = await fetch(`${BASE}/api/generate`, { method: "POST", headers: auth, body: form });
if (submit.status === 402) throw new Error("Not enough credits, top up first");
if (submit.status === 429) throw new Error("Rate limited, retry in a few seconds");
const { id } = await submit.json();
// 2) poll every 4 s until it finishes (typically 45-100 s)
let job;
do {
await sleep(4000);
job = await (await fetch(`${BASE}/api/task/${id}`, { headers: auth })).json();
} while (job.status === "queued" || job.status === "processing");
if (job.status !== "succeeded") throw new Error(`failed (not charged): ${job.error}`);
// 3) download (url is a site path such as /outputs/<id>.png)
const img = await fetch(BASE + job.url, { headers: auth });
await writeFile("image.png", Buffer.from(await img.arrayBuffer()));
console.log("saved image.png, price", job.price, "CNY");
Status values are queued, processing, succeeded and failed. Poll every 3 to 5 seconds. A failed task is not charged and its error message is in error. Error codes to handle: 402 (not enough credits, body has need and have) and 429 (rate limit, back off).
Repeat the refs field for up to 14 images, 8 MB each. More references make generation slower.
curl -X POST https://72agi.com/api/generate \ -H "Authorization: Bearer $VELIN_API_KEY" \ -F "prompt=turn this into a watercolour" \ -F "model=gpt-image-2.5-sunburst" \ -F "refs=@./source.png"
Yes. The standard models cost ¥0.25 (about $0.037) per image at 1K, 2K and 4K. The Pro+ and 2+ variants cost ¥0.35 (about $0.052) at every resolution.
No. Credits are reserved when you submit a task and refunded automatically if it fails.
Yes. Repeat the refs field to send up to 14 reference images (8 MB each) for image-to-image, style and composition work.
Typically about 45 to 100 seconds depending on the model; 4K and many reference images take longer. The API is asynchronous: submit a task, then poll for the result.
Top up with USDT on the billing page; the balance is credited automatically. Card payments are coming soon.
Free tests are available on request: message @ayan8866e on Telegram. There are no automatic free credits.
Yes, on request via Telegram @ayan8866e. Capacity is up to 100,000 images per day.
gpt-image-2.5-flare is the fastest 2.5 model (about 45 seconds) and a good default. gpt-image-2.5-sunburst gives the highest detail (about 100 seconds). gpt-image-2 is the previous generation with a wide choice of aspect ratios.
Want a free test, higher rate limits or volume pricing? Message @ayan8866e on Telegram. Free tests are on request; there are no automatic free credits.
Nano Banana and Gemini are trademarks of Google. GPT Image is a trademark of OpenAI. VELIN is an independent service and is not affiliated with or endorsed by either company. Competitor prices are quoted from the linked public pages on the date shown and may change.