Video Generation API

• Production: https://batch.pipelet.net

Endpoints (quick reference)

• Create request: POST /queue/:modelId
• Check status: GET /queue/:modelId/requests/:requestId/status
• Cancel request: PUT /queue/:modelId/requests/:requestId/cancel
• Fetch result: GET /queue/:modelId/requests/:requestId

Supported modelIds

• Text-to-Video: wan22-fast-t2v, wan22-fast-t2v-premium, wan22-fast-t2v-pro
• Image-to-Video: wan22-fast-i2v, wan22-fast-i2v-premium, wan22-fast-i2v-pro

Request Format

1) Create a request

POST https://batch.pipelet.net/queue/:modelId

Headers:

• Content-Type: application/json
• Authorization: Bearer {your-bearer-token}

Body:

• For text-to-video models, provide a prompt field with text prompts.
• For image-to-video models, provide a prompt and a data_uri field with base64 encoded image.
• Extra parameters are optional, duration for video length in seconds, width and height for video resolution.

Supported Resolution:

• For text-to-video models, 1280x720, 720x1280, 480x832, 832x480, 512x512, 768x768
• For image-to-video models, only height is used, and we only support 720, 480 as input
• For both models, maximum duration for 720p video is 5 seconds. For 480p video it can be 7 seconds.

Example (model wan22-fast-i2v):

curl 'https://batch.pipelet.net/queue/wan22-fast-i2v' \
  -H 'Content-Type: application/json' \
  -d '{"prompt": "A cat wearing a superman cape playing with a dog", "data_uri": "(base64 encoded image)", "duration": 5, "height": 720}'

NOTE: to create a base64 encoded image, do something like base64 -i image.jpg

Example response:

{
  "request_id": "12",
  "response_url": "https://batch.pipelet.net/queue/wan22-fast-i2v/requests/12",
  "status_url": "https://batch.pipelet.net/queue/wan22-fast-i2v/requests/12/status",
  "cancel_url": "https://batch.pipelet.net/queue/wan22-fast-i2v/requests/12/cancel"
}

Notes:

• request_id is your handle for this job for future requests
• Use status_url to poll for progress.
• Use response_url to fetch the completed result.
• Use cancel_url to cancel if the job has not started yet.

2) Poll for status

GET https://batch.pipelet.net/queue/:modelId/requests/:requestId/status

Example:

curl https://batch.pipelet.net/queue/wan22-fast-i2v/requests/12/status

Possible responses:

• In progress

{
  "status": "IN_PROGRESS",
  "statusMessages": {
    "message": "PROCESSING",
    "detail_message": "Progress: 4/24 Tasks done"
  },
  "response_url": "https://batch.pipelet.net/queue/wan22-fast-i2v/requests/12"
}

NOTE: Once a video starts generating, it usually takes 90 seconds to complete for 480p video. For 720p video it takes 180 seconds.

• In queue (not started yet). You will also see the current queue_position, starting at index 0: It means how many jobs are in front of you in the queue.

{
  "status": "IN_QUEUE",
  "queue_position": 1,
  "statusMessages": {},
  "response_url": "https://batch.pipelet.net/queue/wan22-fast-i2v/requests/16"
}

• Other terminal states: COMPLETED, FAILED, CANCELLED. NOTE: always check statusMessages for details.

2) Cancel a queued request

PUT https://batch.pipelet.net/queue/:modelId/requests/:requestId/cancel

• Right now it is only possible to cancel while status is IN_QUEUE (not started).

Example:

curl -XPUT https://batch.pipelet.net/queue/wan22-fast-i2v/requests/16/cancel

If successful, a 204 No Content response is returned; otherwise a conflict (409) is returned if it has already started.

3) Fetch the result

GET https://batch.pipelet.net/queue/:modelId/requests/:requestId

When the job status becomes COMPLETED, fetch the result from the response_url:

curl -XGET https://batch.pipelet.net/queue/wan22-fast-i2v/requests/13

Typical response body:

{
  "video": {
    "data_uri": "(base64_encoded mp4 file content like 4AAQkZJRgABAxxx)"
  }
}

• JSONvideo.data_uri is a base64-encoded string of an MP4 file encoded with H.264.
• Decode and save it locally to obtain the final .mp4 file

Example to save locally (macOS/Linux):

curl -s https://batch.pipelet.net/queue/wan22-fast-i2v/requests/13 \
  | jq -r '.video.data_uri' | base64 --decode > output.mp4

Extra Note for backwards compatibility

All the previous pipelines will produce base64-encoded video files in the response body directly. If you wish to swtich to the S3-presigned URL, add a HTTP header x-pipelet-output: binary to the create request, and the result will be a S3-presigned URL instead.