minimax-h3 视频生成 API

快速信息

获取 API Key

登录 DeepSeek Tavern 控制台，在「令牌管理」中创建 API Key。详细步骤参考：API 密钥获取指南。

Base URL 统一为 https://deepseektavern.com/v1，认证方式：Authorization: Bearer <你的API密钥>。下文示例中的 YOUR_API_KEY 请替换为你自己的 Key。

调用流程

1. POST /v1/videos 提交生成任务，返回任务 id
2. GET /v1/videos/{id} 轮询任务状态，建议每 10 秒一次，直到 status 为 completed
3. 用返回体中的 url 字段下载视频

三种生成模式

请求参数

POST https://deepseektavern.com/v1/videos，JSON 请求体：

curl 示例

文生视频

curl https://deepseektavern.com/v1/videos \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "minimax-h3",
    "prompt": "雨夜的古风酒馆门口，灯笼摇晃，一位披斗篷的剑客推门而入，雨声和风铃声",
    "seconds": "6",
    "resolution": "768P",
    "aspect_ratio": "16:9"
  }'

首尾帧

curl https://deepseektavern.com/v1/videos \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "minimax-h3",
    "mode": "shouweizhen",
    "prompt": "少女从坐着看书到起身走向窗边，镜头平滑跟随",
    "images": [
      "https://example.com/first-frame.png",
      "https://example.com/last-frame.png"
    ],
    "seconds": "5",
    "resolution": "768P",
    "aspect_ratio": "9:16"
  }'

参考生

curl https://deepseektavern.com/v1/videos \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "minimax-h3",
    "mode": "cankaosheng",
    "prompt": "参考图中的角色在参考视频的舞台上跳舞，背景音乐使用参考音频的风格",
    "image_url": ["https://example.com/character.png"],
    "video_url": ["https://example.com/stage-5s.mp4"],
    "audio_url": ["https://example.com/bgm.mp3"],
    "seconds": "8",
    "resolution": "768P",
    "aspect_ratio": "16:9"
  }'

上面这个请求的费用：生成 8 秒 + 参考视频 5 秒 = 13 秒 × 0.5 元 = 6.5 元。

查询任务

curl https://deepseektavern.com/v1/videos/video_xxxxxxxx \
  -H "Authorization: Bearer YOUR_API_KEY"

完整 Python 示例：提交 + 轮询 + 下载

import base64
import time
import requests

API_KEY = "YOUR_API_KEY"
BASE = "https://deepseektavern.com/v1"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}


def to_data_uri(path):
    """本地图片转 data URI（视频不支持，必须传 URL）"""
    mime = "image/png" if path.lower().endswith(".png") else "image/jpeg"
    with open(path, "rb") as f:
        return f"data:{mime};base64," + base64.b64encode(f.read()).decode()


# 1. 提交任务（这里以首尾帧为例；文生视频去掉 mode 和 images 即可）
payload = {
    "model": "minimax-h3",
    "mode": "shouweizhen",
    "prompt": "花苞在晨光中缓缓绽放，露珠滑落，鸟鸣声",
    "images": [to_data_uri("bud.png"), to_data_uri("bloom.png")],
    "seconds": "6",           # 按秒计费：6 秒 = 3 元
    "resolution": "768P",
    "aspect_ratio": "1:1",
}
r = requests.post(f"{BASE}/videos", json=payload, headers=HEADERS, timeout=120)
r.raise_for_status()
video_id = r.json()["id"]
print("任务已提交:", video_id)

# 2. 轮询：每 10 秒一次，总超时 15 分钟（实测 4 秒 768P 约 7 分钟出片）
deadline = time.time() + 15 * 60
while True:
    time.sleep(10)
    info = requests.get(f"{BASE}/videos/{video_id}", headers=HEADERS, timeout=60).json()
    status = info.get("status")
    print("状态:", status, info.get("progress", ""))
    if status == "completed":
        break
    if status in ("failed", "cancelled", "error"):
        raise RuntimeError(f"生成失败: {info}")
    if time.time() > deadline:
        raise TimeoutError("等待超过 15 分钟，可稍后用同一个 id 继续查询")

# 3. 用 url 字段下载（可能是 http 链接，下载后再使用，不要直接嵌入 https 网页）
video_url = info["url"]
with requests.get(video_url, stream=True, timeout=600) as resp:
    resp.raise_for_status()
    with open(f"{video_id}.mp4", "wb") as f:
        for chunk in resp.iter_content(1 << 16):
            f.write(chunk)
print("已保存:", f"{video_id}.mp4")

计费说明