Category: provider

Model Studio VideoRetalk#

Validation#

bash

mkdir -p output/aliyun-videoretalk
python -m py_compile skills/ai/video/aliyun-videoretalk/scripts/prepare_retalk_request.py && echo "py_compile_ok" > output/aliyun-videoretalk/validate.txt

Pass criteria: command exits 0 and output/aliyun-videoretalk/validate.txt is generated.

Output And Evidence#

• Save normalized request payloads, target face selection settings, and task polling snapshots under output/aliyun-videoretalk/.
• Record the exact video/audio input URLs and whether video_extension was enabled.

Use VideoRetalk when the input is already a person video and the job is to replace lip sync with a new speech track.

Critical model names#

Use this exact model string:

• videoretalk

Prerequisites#

• This model currently only supports China mainland (Beijing).
• API is HTTP async only; there is no online console experience.
• Set DASHSCOPE_API_KEY in your environment, or add dashscope_api_key to ~/.alibabacloud/credentials.

Normalized interface (video.retalk)#

Request#

• model (string, optional): default videoretalk
• video_url (string, required)
• audio_url (string, required)
• ref_image_url (string, optional): target face when input video contains multiple faces
• video_extension (bool, optional): extend video to match longer audio
• query_face_threshold (int, optional): 120 to 200

Response#

• task_id (string)
• task_status (string)
• video_url (string, when finished)
• usage (object, optional)

Endpoint and execution model#

• Submit task: POST https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis/
• Poll task: GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}
• HTTP calls are async only and must set header X-DashScope-Async: enable.

Quick start#

bash

python skills/ai/video/aliyun-videoretalk/scripts/prepare_retalk_request.py \
  --video-url "https://example.com/talking-head.mp4" \
  --audio-url "https://example.com/new-voice.wav" \
  --video-extension

Operational guidance#

• Keep input videos front-facing and close enough for stable face tracking.
• If the video contains multiple faces, provide ref_image_url to anchor the intended target.
• If the new audio is longer than the input video, decide explicitly whether to extend the picture track or truncate the audio.
• URLs must be public HTTP/HTTPS links; local file paths are not accepted by the API.

Output location#

• Default output: output/aliyun-videoretalk/request.json
• Override base dir with OUTPUT_DIR.

References#

• references/sources.md