medeo-video（梅迪奥视频）

Medeo Video Generator Skill

Generate AI videos from text. Medeo is an AI video agent that handles full storylines, multi-scene narratives, and screenplays in a single call — shot composition, transitions, pacing, and music are all automatic.

⚠️ Do NOT split stories into multiple calls. Pass the entire screenplay in one --message.

0. Pre-Flight Check (MANDATORY — run before anything else)

Before running any command, check if API key is configured:

python3 {baseDir}/scripts/medeo_video.py config 2>/dev/null | python3 -c "import sys,json; d=json.load(sys.stdin); print('ok' if d.get('apiKey') else 'missing')"

• Output ok → proceed normally
• Output missing (or command fails) → stop immediately, do NOT run any other commands. Send the setup message using the channel-appropriate method:

Feishu — use Feishu API directly (NOT message tool — it won't render cards):

import json, urllib.request
cfg = json.loads(open("/home/ec2-user/.openclaw/openclaw.json").read())
feishu = cfg["channels"]["feishu"]["accounts"]["default"]
token = json.loads(urllib.request.urlopen(urllib.request.Request(
    "https://open.feishu.cn/open-apis/auth/v3/tenant_access_token/internal",
    data=json.dumps({"app_id": feishu["appId"], "app_secret": feishu["appSecret"]}).encode(),
    headers={"Content-Type": "application/json"}
)).read())["tenant_access_token"]
card = {
    "config": {"wide_screen_mode": True},
    "header": {"title": {"tag": "plain_text", "content": "🎬 Video Generation — API Key Required"}, "template": "blue"},
    "elements": [{"tag": "div", "text": {"tag": "lark_md", "content": "You need a **Medeo API Key** to generate videos.\
\
**Steps:**\
1. Go to https://medeo.app/dev/apikey\
   - No account? You'll be guided to sign up. The key appears after login.\
2. Copy the key (starts with `mk_`) and send it back to me.\
\
Once I have it, I'll configure everything for you."}}],
}
urllib.request.urlopen(urllib.request.Request(
    "https://open.feishu.cn/open-apis/im/v1/messages?receive_id_type=open_id",
    data=json.dumps({"receive_id": "<USER_OPEN_ID>", "msg_type": "interactive", "content": json.dumps(card)}).encode(),
    headers={"Authorization": f"Bearer {token}", "Content-Type": "application/json"}
))

Telegram / Discord / Other channels — send plain text via message tool (these channels support markdown natively):

🎬 Video Generation — API Key Required

Steps:
1. Go to https://medeo.app/dev/apikey (sign up if needed — the key appears after login)
2. Copy the key (starts with mk_) and send it back to me

Once I have it, I'll configure everything for you.

Once they provide the key: python3 {baseDir}/scripts/medeo_video.py config-init --api-key "mk_..."

1. First-Time Setup

If no API Key is configured, the script outputs "setup_required": true.

1. Send the user this exact link: https://medeo.app/dev/apikey (this page auto-prompts registration if not logged in, then shows the API key)
2. Once they provide the key: python3 {baseDir}/scripts/medeo_video.py config-init --api-key "mk_..."

2. Generate a Video (5-30 min, always async)

Users only need to know 3 ways to generate a video:

1. Send text → generate video
2. Send text + upload image → generate video using their image
3. Send text + image URL → generate video using the URL image

The agent handles everything else silently.

IMPORTANT: Before spawning the generation task, immediately reply to the user with an acknowledgment like: "🎬 Starting video generation — I'll send you the result in about 5–10 minutes." Do NOT wait in silence. The user should know their request was received.

Usage 1: Text only

python3 {baseDir}/scripts/medeo_video.py spawn-task \
  --message "user's video description" \
  --deliver-to "oc_xxx" \
  --deliver-channel "feishu"

Usage 2: Text + uploaded image (user sends image in chat)

# First: upload-file to get media_id (see Section 3)
python3 {baseDir}/scripts/medeo_video.py spawn-task \
  --message "user's video description" \
  --media-ids "media_01..." \
  --asset-sources my_uploaded_assets \
  --deliver-to "oc_xxx" \
  --deliver-channel "feishu"

Usage 3: Text + image URL

python3 {baseDir}/scripts/medeo_video.py spawn-task \
  --message "user's video description" \
  --media-urls "https://example.com/photo.jpg" \
  --asset-sources my_uploaded_assets \
  --deliver-to "oc_xxx" \
  --deliver-channel "feishu"

Agent auto-behavior: When the user provides images (Usage 2 or 3), always pass --asset-sources my_uploaded_assets so Medeo uses their images instead of generating new ones. The user does not need to know this flag exists.

Internal Parameters (agent use only — never expose to users)

These are handled automatically by the agent. Do NOT mention them to users or ask users to provide them.

Delivery Target (--deliver-to)

This is critical — determines where the generated video gets sent.

How to determine group vs private on Feishu:

• Check is_group_chat in the inbound metadata
• If true → use conversation_label / chat_id (the oc_ value)
• If false → use sender_id (the ou_ value)

Step 2: Use sessions_spawn with the returned args (label: "medeo: <brief>", runTimeoutSeconds: 2400). Step 3: Tell user it's generating. Sub-agent auto-announces when done.

3. Upload Assets

3a. From URL (image already has a public URL)

python3 {baseDir}/scripts/medeo_video.py upload \
  --url "https://example.com/photo.jpg" \
  --project-id "project_01..."          # optional: associate media with existing project
  --no-wait                             # optional: return job_id immediately without polling

3b. From IM attachment (user sends image directly) ← NEW

Use upload-file when the user sends an image via Telegram, Discord, Feishu, or as a local file. This uses the direct upload API (prepare → S3 presigned PUT → register) instead of URL-based upload.

Trigger: Only when the user explicitly requests video generation AND sends an image attachment in the same message (e.g. "make a video with this photo"). Do NOT auto-upload on every image message — other skills or conversations may involve images unrelated to video generation.

# From local file (downloaded by OpenClaw from attachment)
python3 {baseDir}/scripts/medeo_video.py upload-file \
  --file /tmp/user_photo.jpg

# From direct URL (Discord CDN, etc.)
python3 {baseDir}/scripts/medeo_video.py upload-file \
  --url "https://cdn.discordapp.com/attachments/..."

# From Telegram (file_id from message.photo[-1].file_id)
# TELEGRAM_BOT_TOKEN must be set as env var — never pass as CLI arg (ps aux leaks it)
TELEGRAM_BOT_TOKEN="$TELEGRAM_BOT_TOKEN" python3 {baseDir}/scripts/medeo_video.py upload-file \
  --telegram-file-id "AgACAgIAAxk..."

# From Feishu (message_id + image_key from message content)
python3 {baseDir}/scripts/medeo_video.py upload-file \
  --feishu-message-id "om_xxx" \
  --feishu-image-key "img_v3_xxx" \
  --feishu-app-token "$FEISHU_APP_TOKEN"

Output: {"media_id": "media_01...", "filename": "photo.jpg"}

Then pass media_id to generation:

python3 {baseDir}/scripts/medeo_video.py spawn-task \
  --message "Create a video featuring this person" \
  --media-ids "media_01..."

Platform-Specific Image Extraction Guide

Note: Discord attachment URLs are public CDN links — --url works directly. All other platforms require authentication to download.

3c. Inline in generate pipeline

# URL-based (existing behavior)
python3 {baseDir}/scripts/medeo_video.py spawn-task \
  --message "Product showcase for this sneaker" \
  --media-urls "https://example.com/front.jpg" "https://example.com/side.jpg"

Supports .jpg, .png, .webp, .mp4, .mov, .gif. Higher resolution + multiple angles = better results.

3d. Check Upload Status

After upload or upload-file, if you need to check the upload job:

python3 {baseDir}/scripts/medeo_video.py upload-status --job-id "job_01..."

Returns media status (processing, completed, failed) and media_id once done.

4. Low-Level Pipeline Commands (agent internal — never expose to users)

These are for agent debugging or manual intervention only. Users should never see these commands.

Pipeline flow:
  spawn-task (recommended, async)
      └── generate (blocking, same pipeline)
              ├── upload (if --media-urls)
              ├── compose → compose-status (poll)
              └── render → render-status (poll)

All commands support --no-wait to return immediately without polling.

5. Browse Recipes

python3 {baseDir}/scripts/medeo_video.py recipes              # list templates
python3 {baseDir}/scripts/medeo_video.py recipes --cursor <c>  # paginate

Use in generation: --recipe-id "recipe_01...". See docs/recipes.md.

6. Quick Commands Reference (for agent, not user-facing)

7. Key Rules

1. Always async — spawn-task + sessions_spawn for generation
2. One call for stories — full storylines in one --message, never split
3. Insufficient credits — share recharge link from error output
4. IM image upload — Only upload images when the user explicitly asks for video generation with that image. Do NOT auto-upload every image message (user may have other skills installed). When triggered: run upload-file first → get media_id → pass to generation via --media-ids. Never ask the user for a URL if they already sent the image.
5. IM-native delivery — After generation, deliver the video using the IM channel's native method (not just a URL). Each channel has a dedicated delivery script:

- Feishu: python3 {baseDir}/scripts/feishu_send_video.py --video /tmp/result.mp4 --to "oc_xxx_or_ou_xxx" --cover-url "<thumbnail_url>" --duration <ms> (use oc_ chat_id for group chats, ou_ open_id for private chats; chat:oc_xxx and user:ou_xxx prefixed forms are also accepted) - Telegram: Download video, then send via telegram_send_video.py (token from env only):

     curl -sL -o /tmp/medeo_result.mp4 "<video_url>"
     TELEGRAM_BOT_TOKEN="$TELEGRAM_BOT_TOKEN" python3 {baseDir}/scripts/telegram_send_video.py \
       --video /tmp/medeo_result.mp4 \
       --to "<chat_id>" \
       --cover-url "<thumbnail_url>" \
       --duration <seconds> \
       --caption "🎬 Video ready!"

- Discord: Use the message tool directly — download the video to /tmp/result.mp4 via curl -sL -o /tmp/result.mp4 "<video_url>", then call message(action="send", channel="discord", target="<channel_id>", message="🎬 Video ready!", filePath="/tmp/result.mp4"). For files >25 MB, send video_url as a plain link instead. - WhatsApp / Signal / Other: Use the message tool with media parameter, or share video_url as a link if native sending is unavailable. - Cover image URL: The generate output JSON includes thumbnail_url — the API always returns this field. Constructed as {ossBaseUrl}/{thumbnail_relative_path} (e.g. https://oss.prd.medeo.app/assets/medias/media_xxx.png). - Video URL: Same pattern — {ossBaseUrl}/{video_relative_path} (e.g. https://oss.prd.medeo.app/exported_video/v_xxx). - Security: Never pass bot tokens as CLI args (visible in ps). Always use env vars: TELEGRAM_BOT_TOKEN, DISCORD_BOT_TOKEN.

1. Timeline completion — Medeo's backend is an AI agent. Generated images/videos must be added to the Timeline to trigger task completion and rendering. Always append to your prompt: "Add the generated video/image to the Timeline."

8. Error Handling

9. Reference Docs

• docs/recipes.md — Full recipe browsing and pagination
• docs/assets-upload.md — All upload paths (URL, local file, IM attachments), platform-specific guides, upload vs upload-file comparison
• docs/feishu-send.md — Sending generated video via Feishu (cover image, duration, compression)
• docs/multi-platform.md — Multi-platform video delivery (Feishu, Telegram, Discord, WhatsApp)

10. Data Storage

All data in ~/.openclaw/workspace/medeo-video/: config.json (API key), last_job.json (latest job), history/ (last 50 jobs).

11. Security Notes

• API key resolution: env var MEDEO_API_KEY → config.json → built-in defaults. No legacy system-level files are read.
• Feishu delivery: feishu_send_video.py reads appId + appSecret from local ~/.openclaw/openclaw.json to call Feishu Open API. Credentials stay local and are never transmitted beyond the Feishu API.
• Telegram delivery: Bot token is read from TELEGRAM_BOT_TOKEN env var only (never CLI args).
• No secrets in skill directory: config.json lives in the runtime data directory (~/.openclaw/workspace/medeo-video/), not in the skill source directory.