Aliyun Wan Digital Human

Use when generating talking, singing, or presentation videos from a single character image and audio with Alibaba Cloud Model Studio digital-human model `wan...

Security Scan

The skill appears to be a simple request-preparer for Alibaba Cloud's 'wan' digital-human models, but its instructions and metadata disagree about credential requirements and promised runtime behaviors that the included code does not perform.

Purpose & Capability

The name/description (generate talking/singing avatar video via Alibaba Cloud Model Studio) matches the included script which only prepares a JSON request payload. However the SKILL.md also demands a DASHSCOPE_API_KEY or an entry in ~/.alibabacloud/credentials even though the repository metadata declares no required env vars and the included script does not read credentials or make network calls. That mismatch is unexpected and unexplained.

Instruction Scope

SKILL.md instructs validating images, running a detection model first, polling tasks, and saving normalized payloads, chosen resolution, and task polling snapshots under output/. The provided script only writes a single request.json and prints its path; it does not perform detection, polling, network calls, or record the additional evidence SKILL.md promises. The instructions also tell users to set DASHSCOPE_API_KEY or a credentials file entry, but no code in this bundle uses those values.

Install Mechanism

This is instruction-only with a tiny helper script and no install spec — minimal disk/runtime footprint and no archived downloads. Low install risk.

Credentials

Registry metadata claims no required env vars, but SKILL.md requires DASHSCOPE_API_KEY (or a dashscope_api_key in ~/.alibabacloud/credentials). Requesting cloud credentials is reasonable for a cloud integration, but the specific variable name and the mismatch with metadata are unexplained. It's unclear whether DASHSCOPE_API_KEY is an Alibaba Cloud credential, a third-party proxy, or something else — this should be clarified before trusting keys.

Persistence & Privilege

The skill does not request always: true, does not include install scripts that modify global agent settings, and is user-invocable only. No elevated persistence or system-wide changes are present.

What to consider before installing

What to check before installing or using this skill: - Expectation vs. code: the included script only creates a request.json and does not call any remote API, perform detection, or poll tasks — SKILL.md describes additional runtime steps that are not implemented here. Ask: where (which component) actually sends requests to Alibaba Cloud and performs detection/polling? Obtain that code or documentation before sending real credentials or data. - Credential confusion: SKILL.md asks for DASHSCOPE_API_KEY or adding dashscope_api_key to ~/.alibabacloud/credentials, but the registry metadata lists no required env vars. Verify what DASHSCOPE_API_KEY is (an Alibaba Cloud key, a Dashscope proxy key, or something else) and only provide least-privileged keys. Prefer IAM/temporary credentials and avoid placing broad keys in shared ~/.alibabacloud/credentials unless you understand the access scope. - Data privacy: the skill processes images and audio which may be sensitive. Confirm which service and endpoint will receive the media (Alibaba Cloud Model Studio in Beijing region is claimed) and that you are comfortable with regional data handling and retention policies. - Operational testing: run the bundled script in a safe environment to confirm it only writes request.json. Do not provide production credentials until you can trace the complete request flow (who/what will read request.json and where it will be sent). - Ask the publisher: request the full implementation that performs detection and submission (or a homepage/repository) and clarify why the SKILL.md promises outputs and behaviors not present in the bundle. The mismatch could be benign (this is only a helper) or indicate missing/misplaced code that will run elsewhere — confirm before trusting credentials or uploading private media.

Category: provider

Model Studio Digital Human

Validation

mkdir -p output/aliyun-wan-digital-human
python -m py_compile skills/ai/video/aliyun-wan-digital-human/scripts/prepare_digital_human_request.py && echo "py_compile_ok" > output/aliyun-wan-digital-human/validate.txt

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

Output And Evidence

• Save normalized request payloads, chosen resolution, and task polling snapshots under output/aliyun-wan-digital-human/.
• Record image/audio URLs and whether the input image passed detection.

Use this skill for image + audio driven speaking, singing, or presenting characters.

Critical model names

Use these exact model strings:

• wan2.2-s2v-detect
• wan2.2-s2v

Selection guidance:

• Run wan2.2-s2v-detect first to validate the image.
• Use wan2.2-s2v for the actual video generation job.

Prerequisites

• China mainland (Beijing) only.
• Set DASHSCOPE_API_KEY in your environment, or add dashscope_api_key to ~/.alibabacloud/credentials.
• Input audio should contain clear speech or singing, and input image should depict a clear subject.

Normalized interface (video.digital_human)

Detect Request

• model (string, optional): default wan2.2-s2v-detect
• image_url (string, required)

Generate Request

• model (string, optional): default wan2.2-s2v
• image_url (string, required)
• audio_url (string, required)
• resolution (string, optional): 480P or 720P
• scenario (string, optional): talk, sing, or perform

Response

• task_id (string)
• task_status (string)
• video_url (string, when finished)

Quick start

python skills/ai/video/aliyun-wan-digital-human/scripts/prepare_digital_human_request.py \
  --image-url "https://example.com/anchor.png" \
  --audio-url "https://example.com/voice.mp3" \
  --resolution 720P \
  --scenario talk

Operational guidance

• Use a portrait, half-body, or full-body image with a clear face and stable framing.
• Match audio length to the desired output duration; the output follows the audio length up to the model limit.
• Keep image and audio as public HTTP/HTTPS URLs.
• If the image fails detection, do not proceed directly to video generation.

Output location

• Default output: output/aliyun-wan-digital-human/request.json
• Override base dir with OUTPUT_DIR.

References

• references/sources.md