视频生成调用指南
通过统一网关调用 Sora 2、可灵 1.6、海螺 2.3、字节 Seedance 等视频生成模型。
模型清单
| 模型 ID | 厂商 | 时长 | 分辨率 |
|---|---|---|---|
sora-2 | OpenAI | 5-60s | 720p / 1080p |
kling-v1.6 | 快手可灵 | 5-10s | 720p |
MiniMax-Hailuo-2.3 | MiniMax | 6s | 720p |
doubao-seedance-1-0-pro | 字节 | 5s | 720p |
视频生成是异步任务
视频生成耗时 1-5 分钟,调用是异步模式:
- 提交任务 → 拿
task_id - 轮询任务状态
- 完成后拿到视频 URL
Python 调用
python
import time, requests
API_KEY = "sk-xxx"
BASE = "http://xdhdancer.top"
# 1. 提交任务
resp = requests.post(
f"{BASE}/v1/videos/generations",
headers={"Authorization": f"Bearer {API_KEY}"},
json={
"model": "sora-2",
"prompt": "a panda eating bamboo in slow motion, cinematic",
"duration": 6,
"size": "1280x720",
},
)
task_id = resp.json()["task_id"]
# 2. 轮询
while True:
status = requests.get(
f"{BASE}/v1/videos/{task_id}",
headers={"Authorization": f"Bearer {API_KEY}"},
).json()
if status["status"] == "succeeded":
print("Video URL:", status["video_url"])
break
elif status["status"] == "failed":
print("Failed:", status["error"])
break
time.sleep(10)
print("...waiting")Prompt 技巧
视频 prompt 需要描述动态:
a fox running through autumn forest, leaves falling,
camera follows from behind, golden hour lighting避免静态描述(如 a fox standing),模型可能生成几乎不动的画面。