API 文档

用程序批量生成图片和视频。API 和网站共用同一个账号、同一份积分，价格与网站一致；任务失败会自动退回积分。

快速开始

https://niuzhatian.com

curl https://niuzhatian.com/v1/balance \
  -H "Authorization: Bearer nzt-你的密钥"

认证

在每个请求的请求头里带上密钥：

Authorization: Bearer nzt-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

调用流程

生成是异步的：

可以一次性并发提交很多任务，不用等上一个完成。

模型列表

返回当前可用的全部模型。模型会随平台更新，请以这个接口的返回为准，不要把模型写死在代码里太久。

{
  "object": "list",
  "data": [
    {
      "id": "m-3f8a1c92d4",
      "name": "Seedance 2.5 VIP（720p）",
      "type": "video",
      "resolutions": ["720p"],
      "durations": { "min": 5, "max": 30 },
      "aspect_ratios": ["16:9", "9:16", "1:1", "4:3", "3:4", "21:9"],
      "references": { "images": 9, "audios": 3, "videos": 3 },
      "price": { "unit": "per_task", "by_resolution": { "720p": 400 } }
    },
    {
      "id": "m-xxxxxxxxxx",
      "name": "某图片模型",
      "type": "image",
      "sizes": [ { "value": "2K", "price": 30 } ],
      "max_reference_images": 4,
      "price": { "unit": "per_image", "by_size": { "2K": 30 } }
    }
  ]
}

上传参考素材

给视频模型用的参考素材要先上传，拿到素材编号后再提交任务。请求体直接放文件内容（不是表单），并设置对应的 Content-Type。

curl "https://niuzhatian.com/v1/media?type=image&name=hero.jpg" \
  -H "Authorization: Bearer nzt-你的密钥" \
  -H "Content-Type: image/jpeg" \
  --data-binary @hero.jpg

返回：

{ "id": "xxxxxxxx:rAbCdEf...", "type": "image" }

提交任务

请求体为 JSON（Content-Type: application/json）。

生成视频

curl https://niuzhatian.com/v1/generations \
  -H "Authorization: Bearer nzt-你的密钥" \
  -H "Content-Type: application/json" \
  -d '{"model":"m-3f8a1c92d4","prompt":"清晨的海边，一只金毛在奔跑，电影感","duration":10,"aspect_ratio":"9:16"}'

带参考素材生成视频

curl https://niuzhatian.com/v1/generations \
  -H "Authorization: Bearer nzt-你的密钥" \
  -H "Content-Type: application/json" \
  -d '{"model":"m-3f8a1c92d4","prompt":"@图片1 是主角，在海边奔跑，背景音乐用 @音频1","duration":10,
       "media":[{"id":"上传返回的图片编号","type":"image"},{"id":"上传返回的音频编号","type":"audio"}]}'

生成图片

curl https://niuzhatian.com/v1/generations \
  -H "Authorization: Bearer nzt-你的密钥" \
  -H "Content-Type: application/json" \
  -d '{"model":"模型id","prompt":"一杯冰美式，白色背景，商业摄影"}'

返回

{
  "id": "tQ2x9vLm4kP7wZ",
  "type": "video",
  "model": "m-3f8a1c92d4",
  "status": "queued",
  "cost": 400
}

查询任务

{
  "id": "tQ2x9vLm4kP7wZ",
  "type": "video",
  "model": "Seedance 2.5 VIP（720p）",
  "status": "succeeded",
  "progress": null,
  "error": null,
  "cost": 400,
  "result_url": "https://niuzhatian.com/v1/generations/tQ2x9vLm4kP7wZ/content",
  "created_at": "2026-11-02T08:00:00.000Z"
}

建议每 5–10 秒查一次。图片一般几十秒，视频一般几分钟到十几分钟。

下载结果

返回图片或视频文件本身，同样需要带 Authorization。加 ?dl=1 时可能会 302 跳转到平台的下载地址，请让你的程序跟随跳转（curl 加 -L）。

curl -L https://niuzhatian.com/v1/generations/tQ2x9vLm4kP7wZ/content \
  -H "Authorization: Bearer nzt-你的密钥" -o result.mp4

查询余额

{ "balance": 12800 }

错误码

出错时返回对应的 HTTP 状态码，响应体为：

{ "error": { "message": "积分不足：这个任务需要 400 积分，当前只有 120 积分，请先到「积分」页兑换" } }

限制与计费

完整示例（Python）

import time, requests

BASE = "https://niuzhatian.com"
H = {"Authorization": "Bearer nzt-你的密钥"}

# 1. 提交
r = requests.post(f"{BASE}/v1/generations", headers=H, json={
    "model": "m-3f8a1c92d4",
    "prompt": "清晨的海边，一只金毛在奔跑，电影感",
    "duration": 10,
    "aspect_ratio": "9:16",
})
r.raise_for_status()
task = r.json()
print("已提交", task["id"], "花费", task["cost"], "积分")

# 2. 等待完成
while True:
    t = requests.get(f"{BASE}/v1/generations/{task['id']}", headers=H).json()
    if t["status"] in ("succeeded", "failed", "canceled"):
        break
    time.sleep(8)

# 3. 下载
if t["status"] == "succeeded":
    data = requests.get(t["result_url"], headers=H).content
    open(task["id"] + ".mp4", "wb").write(data)
    print("已保存")
else:
    print("失败：", t["error"], "（积分已退回）")

带参考图生成视频（Python）

def upload(path, kind, ctype):
    with open(path, "rb") as f:
        r = requests.post(f"{BASE}/v1/media", params={"type": kind, "name": path},
                          headers={**H, "Content-Type": ctype}, data=f.read())
    r.raise_for_status()
    return r.json()          # {"id": "...", "type": "image"}

hero = upload("hero.jpg", "image", "image/jpeg")
r = requests.post(f"{BASE}/v1/generations", headers=H, json={
    "model": "m-3f8a1c92d4",
    "prompt": "@图片1 是主角，在雨夜的街头回头微笑",
    "duration": 10,
    "media": [hero],
})
print(r.json())

批量并发提交

from concurrent.futures import ThreadPoolExecutor

prompts = ["海边日出", "雪山日落", "城市夜景"]  # 可以是几百条

def submit(p):
    r = requests.post(f"{BASE}/v1/generations", headers=H,
                      json={"model": "m-3f8a1c92d4", "prompt": p, "duration": 5})
    return r.json()

with ThreadPoolExecutor(20) as ex:
    tasks = list(ex.map(submit, prompts))
print([t.get("id") or t.get("error") for t in tasks])