mdclaw多模态图片视频生成

MDClaw OpenClaw API 技能，支持文字转语音(TTS)、文生图(Text to Image)、文生视频(Text to Video)、图生视频(Image to Video)等多模态 AI 能力。通过网关服务统一调用，支持账号注册、图片上传、任务轮询等完整功能。

Security Scan

The skill appears to implement a multimodal client that sends prompts, files, and API keys to an external gateway — the code and SKILL.md are coherent with that purpose, but registry metadata omits the required API key and the gateway/owner are unknown, so you should be cautious before installing or uploading sensitive files.

Purpose & Capability

Name/description, SKILL.md, README, and mdclaw_client.py are consistent: a client for a remote MDClaw gateway providing TTS, text->image, text->video, image->video, upload, search, etc. However the published registry metadata (in the evaluation header) stated 'Required env vars: none' and 'Primary credential: none' while the code and clawhub.toml explicitly require MDCLAW_API_KEY. This mismatch is an incoherence in metadata vs. actual capability requirements.

Instruction Scope

SKILL.md and the client instruct the agent to call a single external gateway (https://backend.appmiaoda.com/.../openclaw-skill-gateway). The client will also read local image files (upload_image) and POST file contents to an upload endpoint derived from the gateway. That behavior is expected for an image-uploading/video-generation client, but it means local files and any prompts are transmitted to an external third party. The SKILL.md does not warn about uploading sensitive files or who controls the gateway; the owner/source/homepage are unknown.

Install Mechanism

No install spec; included files are Python source and examples. Dependency is only requests (requirements.txt). Nothing is downloaded from arbitrary URLs during install. This is low install risk.

Credentials

The code and clawhub.toml require an MDCLAW_API_KEY (passed via X-API-Key) which is proportional to the stated purpose. However the registry metadata provided at the top of this evaluation incorrectly lists no required env vars/primary credential — an inconsistency that could mislead users or automated installers into not providing/validating credentials. Requesting an API key is expected; requesting additional unrelated secrets would be concerning (not present here).

Persistence & Privilege

The skill does not request always:true, does not modify other skills, and has no install-time persistence mechanism beyond the code files themselves. Normal autonomous invocation is allowed (default) but not combined with any other elevated privilege.

What to consider before installing

Key points to consider before installing: - The client will send your prompts, API key, and any uploaded files to the external gateway at backend.appmiaoda.com. Do not upload sensitive images or data unless you trust that endpoint and its operator. - The code and clawhub.toml require MDCLAW_API_KEY, but the registry header metadata omitted that—this mismatch is a red flag. Verify credentials handling and update metadata before trusting automated installs. - The package has no listed homepage or identifiable owner; if you need this skill, ask the publisher for provenance (who runs the gateway, privacy/data-retention policies, and SLA). - If you want to test it, run in an isolated environment or sandbox with a throwaway API key and avoid uploading private files. Monitor egress traffic to confirm only expected endpoints are contacted. - If you cannot confirm the gateway operator or metadata, prefer alternative well-known providers or self-hosted solutions. If you proceed, grant the minimum necessary permissions to the API key and rotate it after testing.

MDClaw 多模态 技能

功能描述

通过 MDClaw OpenClaw API 网关访问多模态 AI 能力：

• 文字转语音 (TTS) - 将文本转换为自然语音，支持多种音色
• 文生图 (Text to Image) - 根据文字描述生成图片，支持多种宽高比和模型
• 文生视频 (Text to Video) - 根据文字描述生成视频（异步任务）
• 图生视频 (Image to Video) - 将图片转换为视频（异步任务）
• 图片上传 - 上传本地图片获取 URL，用于图生视频
• 全网搜索 - AI 驱动的全网搜索
• 天气查询 / 网页总结 - 实用辅助功能

认证方式

API Key 是主要认证方式，通过 X-API-Key 请求头传递。

获取 API Key 的方式：

client = MDClawClient()
result = client.agent_register("用户名", "密码")
api_key = result["result"]["api_key"]

或设置环境变量：

export MDCLAW_API_KEY="你的API Key"

API 参考

请求格式

POST https://backend.appmiaoda.com/projects/supabase287606411725160448/functions/v1/openclaw-skill-gateway

Headers:
  X-API-Key: 你的API Key
  Content-Type: application/json

Body:
  {"skill_id": "技能名称", "parameters": {...}}

技能列表

技能	参数	说明
text_to_speech	text, model?, voice_id?	文字转语音
text_to_image	prompt, model?, aspect_ratio?, n?	文生图
text_to_video	prompt, model?, duration?	文生视频（异步）
image_to_video	image, prompt?, model?, duration?	图生视频（异步）
video_status	task_id	查询视频任务状态
ai_search	query	全网搜索
weather_query	city	天气查询
web_summary	url	网页总结

视频生成说明

视频生成是异步操作：

1. 调用 text_to_video 或 image_to_video 获取 task_id
2. 用 video_status(task_id) 轮询状态
3. 状态流转: Preparing → Queueing → Processing → Success / Fail

注意: 不要传递 resolution 参数，否则 API 不返回 task_id。

使用示例

from mdclaw_client import MDClawClient

# 初始化（从环境变量读取 API Key）
client = MDClawClient()

# 文字转语音
result = client.text_to_speech("你好，这是语音测试")
audio_url = result["result"]["audio_url"]

# 文生图
result = client.text_to_image("一只可爱的橘猫在阳光下伸懒腰", aspect_ratio="9:16")
image_url = result["result"]["image_urls"][0]

# 文生视频（异步）
result = client.text_to_video("一只金毛犬在公园快乐地奔跑", duration=6)
task_id = result["result"]["task_id"]

# 等待视频完成
video_result = client.wait_for_video(task_id)
video_url = video_result["result"]["url"]

# 上传图片并生成视频
upload = client.upload_image("local_image.jpg")
image_url = upload["result"]["url"]
result = client.image_to_video(image_url, "让画面动起来")

错误处理

result = client.text_to_image("prompt")

if not result.get("success"):
    print(f"错误: {result.get('error')}")

依赖

requests>=2.31.0

---

版本: v2.0.0 更新日期: 2026-03-24