majiayu000 Sora Video Generation
Generate AI videos through AceDataCloud's OpenAI Sora API.
Authentication
export ACEDATACLOUD_API_TOKEN="your-token-here"
Quick Start
curl -X POST https://api.acedata.cloud/sora/videos \
-H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{"prompt": "a golden retriever running on a beach at sunset", "model": "sora-2", "wait": true}'
Models
| Model | Duration | Quality | Best For |
|---|---|---|---|
sora-2 |
10–15s | Standard | Most tasks (default) |
sora-2-pro |
10–25s | Higher | Premium quality, longer videos |
Workflows
1. Text-to-Video
POST /sora/videos
{
"prompt": "a busy Tokyo street at night with neon signs reflecting in rain puddles",
"model": "sora-2",
"size": "small",
"duration": 10,
"orientation": "landscape"
}
2. Image-to-Video
Use reference images to guide generation.
POST /sora/videos
{
"prompt": "the scene gradually comes alive with gentle motion",
"image_urls": ["https://example.com/scene.jpg"],
"model": "sora-2",
"orientation": "landscape"
}
3. Character-Driven Video
Extract a character from an existing video and use them in a new scene.
POST /sora/videos
{
"prompt": "the character walks through a futuristic city",
"character_url": "https://example.com/source-video.mp4",
"character_start": 2.0,
"character_end": 5.0,
"model": "sora-2-pro"
}
Parameters
| Parameter | Values | Description |
|---|---|---|
size |
"small", "large" |
Video resolution |
duration |
10, 15, 25 |
Duration in seconds (25 only with sora-2-pro) |
orientation |
"landscape" (16:9), "portrait" (9:16), "square" (1:1) |
Video orientation |
Task Polling
POST /sora/tasks
{"task_id": "your-task-id"}
States: pending → succeeded or failed.
MCP Server
pip install mcp-sora
Or hosted: https://sora.mcp.acedata.cloud/mcp
Key tools: sora_generate_video, sora_generate_video_from_image, sora_generate_video_with_character
Gotchas
- Duration of 25 seconds is only available with
sora-2-promodel size: "large"produces higher resolution but costs more and takes longer- Character-driven generation requires
character_startandcharacter_endtimestamps (in seconds) from the source video orientationsets the aspect ratio — use"portrait"for mobile-first content- Task states use
"succeeded"(not "completed") — check for this value when polling