Official GPT direct · per token (#184 / #185)

1Groups and models

Both groups take the same requests; switching is just a different key. Group names: #185 「【GPT 官方直连 按量】OAI · Image 2 / Flare / Sunburst」, #184 「【GPT 官方直连 按量】Azure · Image 2」.

2Billing

Both groups bill per token: unit price = OpenAI list price × group multiplier (currently 2.5; see the model plaza):

Measured average charge per image (2026-10-27/28, #185 gpt-image-2.5-flare / sunburst, 16:9 text-to-image, multiplier 2.5):

• gpt-image-2 1024×1024 low is about $0.015 per image; #184 at 3840×2160 low measured $0.028.
• Edits add image-input tokens: one 512×512 reference is about 1024 tokens (about $0.02). Measured 2K high edit about $0.137, 4K high edit about $0.28.
• Failed requests (HTTP 4xx / 5xx) are not charged.

3Quick start

Create an API key for group #185 or #184 in the dashboard. Text to image:

curl https://yyds.seqnode.com/v1/images/generations \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-sunburst",
    "prompt": "A quiet lakeside town at dawn, soft mist",
    "size": "2048x1152",
    "quality": "high",
    "n": 1,
    "response_format": "url"
  }'

Edit with a reference image (multipart, field image):

curl https://yyds.seqnode.com/v1/images/edits \
  -H "Authorization: Bearer $API_KEY" \
  -F model=gpt-image-2.5-flare \
  -F "prompt=Turn this photo into a watercolor painting" \
  -F size=2048x1152 \
  -F quality=high \
  -F response_format=url \
  -F "image=@input.png"

Python (handles both link and Base64 responses):

import base64, requests

API_KEY = "sk-..."  # 分组 #185 或 #184 的 Key / a key of group #185 or #184
r = requests.post(
    "https://yyds.seqnode.com/v1/images/generations",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={"model": "gpt-image-2", "prompt": "a red fox in snow",
          "size": "1024x1024", "quality": "low", "response_format": "url"},
    timeout=600,
)
r.raise_for_status()
d = r.json()
item = d["data"][0]
if item.get("url"):
    open("out.png", "wb").write(requests.get(item["url"], timeout=300).content)
else:
    open("out.png", "wb").write(base64.b64decode(item["b64_json"]))
print(d.get("usage"))

4Try itCharged for real

Call this site's API straight from the browser. The key goes only to this site; it is kept in this browser only if you tick «Remember».

5Request parameters

6Response

Generations and edits return the same shape:

{
  "created": 1793151600,
  "data": [{ "url": "https://img.yydsbest.com/adb085ce-….png" }],
  "usage": {
    "input_tokens": 16,
    "input_tokens_details": { "text_tokens": 16, "image_tokens": 0 },
    "output_tokens": 371,
    "output_tokens_details": { "image_tokens": 371, "text_tokens": 0 },
    "total_tokens": 387
  }
}

7Errors

Failed requests are not charged. Error bodies are {"error": {"type", "message", "request_id"}} (key check failures: {"code", "message"}); include the request_id when reporting a problem.

8Latency and network

• Measured generation time: 1K about 20–35 s, 2K about 20–50 s, 4K about 40–70 s; max up to about 100 s. Use a client timeout of at least 300 s.
• 4K Base64 responses are large (15–26 MB) and still have to be downloaded after the image is ready. At peak hours some mainland China networks download from this site at tens of KB/s; one 4K Base64 response took over 20 minutes. With url the API returns almost at once and the PNG, about a quarter smaller than Base64, downloads separately.
• For large batches, limit the number of requests in flight so downloads do not compete for bandwidth.

9Related guides

• Image API reference
• Web GPT (#75 / #159, per image)
• Full-parameter Image 2.5 4K
• C2PA signature check (live, Chinese)