Videos

Use https://openai.2yanx.dpdns.org with Authorization: Bearer YOUR_API_KEY. Use a customer key from the console, not an administrator token. Production clients should use asynchronous tasks or Responses SSE for long video requests.

Model capabilities

Model	Text	Single image	Multiple references	Duration	Default size
cogvideo-fast	Yes	Yes	Yes	Text/image 1–10s, reference 3–10s; default 5s	1920×1080
cogvideo-pro	Yes	Yes	Yes	Text/image 1–10s, reference 3–10s; default 5s	1920×1080
cogvideo-basic	Yes	Yes	No	1–5s; default 5s	1920×1080
cogvideo-short	Yes	No	No	Exactly 8s	720×1280

These are site product aliases. The cogvideo prefix does not assert generation with open-source CogVideo weights.

Default-group pricing is USD 0.04109589 per successful video, approximately CNY 0.30 at the site's current conversion. Each request produces one video; duration, resolution, and generation mode do not add a multiplier. Submission reserves credit, failed tasks are refunded, and invalid parameters do not charge. Other groups follow current console pricing and group multipliers.

Create, poll, and download

curl https://openai.2yanx.dpdns.org/v1/videos \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"model":"cogvideo-fast","prompt":"A red paper boat drifting on a quiet pond","seconds":"5","size":"1920x1080"}'

{"id":"task_EXAMPLE","object":"video","status":"queued","progress":0,"created_at":1790980000}

Save id. Poll every 3–5 seconds with a customer key belonging to the task owner:

curl https://openai.2yanx.dpdns.org/v1/videos/task_EXAMPLE \
  -H 'Authorization: Bearer YOUR_API_KEY'

Normal states are queued, in_progress, completed, and failed. Failed tasks carry video_generation_failed. An unrecognized task state may appear as unknown; it does not indicate completion. This endpoint returns status, not a data[0].url download result.

After completed, download the video:

curl https://openai.2yanx.dpdns.org/v1/videos/task_EXAMPLE/content \
  -H 'Authorization: Bearer YOUR_API_KEY' -o video.mp4

The content endpoint supports HEAD and byte Range requests.

Request parameters

model and a non-blank prompt (1–4096 characters) are required. seconds accepts an integer or integer string, within the chosen model's limits. Fast/Pro/Basic sizes: 1920x1080, 1080x1920, 1080x1080, 1440x1080, 1080x1440; Short only accepts 720x1280. Read actual encoded dimensions from the downloaded file.

Use input_reference for a single image URL, image data URL, or multipart file. Multipart accepts one image file up to 20 MiB; overall request limits also apply. Short does not support image inputs.

Multiple references are available on Fast/Pro only:

{
  "model": "cogvideo-pro",
  "prompt": "The red paper boat from [@图1] and the yellow boat from [@图2] drift side by side",
  "seconds": "5",
  "metadata": {
    "mode": "reference_to_video",
    "reference_images": ["https://example.com/red.png", "https://example.com/yellow.png"],
    "generation_mode": "pro_ad"
  }
}

metadata.mode, when present, must match supplied inputs: text_to_video, image_to_video, or reference_to_video. Reference mode uses speed, pro, or pro_ad (default) and requires 3–10 seconds. Reference order controls the [@图N] prompt references.

Compatibility aliases: duration means seconds; image_url means input_reference; reference_images and generation_mode can appear at top level; resolution/aspect_ratio can replace size and may also appear in metadata. Duplicate representations must agree. Single-image and reference-image inputs cannot be mixed. n and sample_count only accept 1. Unknown fields are rejected. Start/end frames, extension, remix, and multiple outputs are unavailable.

Stream progress with Responses

POST /v1/responses uses input for text and the same video parameters:

curl -N https://openai.2yanx.dpdns.org/v1/responses \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{"model":"cogvideo-fast","input":"A red paper boat drifting on a pond","seconds":"5","stream":true}'

The SSE stream reports progress, outputs an HTML video element with the final link, and terminates with response.completed. It does not stream unfinished video frames. Check the final response state and output before treating the request as successful.

Structured input also accepts message content with text and images:

{
  "model": "cogvideo-fast",
  "stream": true,
  "input": [{"role":"user","content":[
    {"type":"input_text","text":"Animate this paper boat on a quiet pond"},
    {"type":"input_image","image_url":"https://example.com/boat.png"}
  ]}]
}

Background Responses

Set background:true to receive a resp_... ID immediately. Poll GET /v1/responses/{response_id}. The response's metadata.task_id can be used with /v1/videos/{task_id}/content.

Successful output[].content[].text contains <video controls src="PRIVATE_SHARE_URL"></video>. Protect this URL like a private file sharing credential. Re-querying may generate a new sharing link. Authenticated content download is also available with the owning customer's key.

Without stream or background, the HTTP request waits for completion. A proxy may disconnect while the backend still completes and charges, so use asynchronous tasks or SSE and investigate the existing task before resubmitting.

These video parameters and outputs are site extensions. Clients supporting only /v1/chat/completions cannot directly use these video models.