hidream-api-gen（hidream API GEN 文档）

HD AIGC Skills

This skill provides per-model scripts that validate parameters and call shared request runners.

Security Disclosure (AI Agents)

• Credential Access: This skill reads HIDREAM_AUTHORIZATION from environment variables or a secure local config file (~/.config/openclaw/hidream_config.json, permission 0600).
• File System Access: Scripts read local image/video files provided in arguments to base64 encode them for API transmission. No other file system access occurs.
• Network Activity: Requests are sent exclusively to the configured HIDREAM_ENDPOINT (default: https://vivago.ai).

Configuration & Usage (AI Agent Instructions)

When a user wants to use this skill, follow these steps:

1. Check for Token: The system automatically checks for the API key in the following order:

* Local Env File (Primary): ~/.openclaw/.env (reads HIDREAM_AUTHORIZATION=sk-...) * Environment Variable: HIDREAM_AUTHORIZATION * Config File: ~/.config/openclaw/hidream_config.json

1. Request Token (If Missing or 401): Instruct the user to update their credentials:

* Recommended: Provide their API Key (starts with sk-) and ask the agent to save it: echo "HIDREAM_AUTHORIZATION=sk-..." > ~/.openclaw/.env * Interactive: Run python3 scripts/configure.py to interactively save the token.

1. Handle Missing Prompts: If the user asks to generate an image or video but does not provide a specific prompt, DO NOT generate a random test image. Instead, politely ask the user what they would like to generate.
2. Generate: Use the Python interface to generate content based on user requests.
3. Save Output: When generation is complete and returns a media URL (image or video), ALWAYS download and save the file to the assets/ directory within this skill's folder (e.g., assets/generated_image.png). Do NOT use ~/.openclaw/workspace/output/ as it may have permission issues. Create the assets/ directory if it does not exist.
4. Present Results: When showing generated images or videos to the user, ALWAYS use the following Markdown formats to ensure proper rendering in the Claw interface. Use the local path if downloaded, or the remote URL:

Error Handling & Troubleshooting

• Resolution too small (ValueError): Seedream M2 requires high resolution (at least 2560*1440 or 2048*2048). If you get this error, increase the resolution parameter.
• 401 Unauthorized (invalid token): The token is invalid or expired. Update it by writing the new token to ~/.openclaw/.env:

    echo "HIDREAM_AUTHORIZATION=sk-..." > ~/.openclaw/.env

• Insufficient Credits (Code 2007): Instruct the user to go to https://vivago.ai/platform/info to recharge credits.

Python Interface (Recommended)

You can call the scripts directly from Python code. This is the preferred way for AI agents to interact.

Image Generation (Seedream)

from scripts.seedream import run as run_seedream

# Example: Generate an image
try:
    result = run_seedream(
        version="M2",
        prompt="A cyberpunk cat on the moon",
        resolution="2048*2048",
        authorization="sk-..." # Optional if env var is set
    )
    print(result)
except Exception as e:
    print(f"Error: {e}")

Video Generation (Kling)

from scripts.kling import run as run_kling

# Example: Generate a video
try:
    result = run_kling(
        version="Q2.5T-std",
        prompt="A cyberpunk cat running on neon streets",
        duration=5,
        authorization="sk-..." # Optional if env var is set
    )
    print(result)
except Exception as e:
    print(f"Error: {e}")

Structure

• scripts/commom/base_image.py: shared OpenClaw image request runner
• scripts/common/base_video.py: shared OpenClaw video request runner
• scripts/common/task_client.py: http request runner
• scripts/*.py: per-model scripts (parameter parsing + payload only)

Auth and Endpoints

Set one of the following environment variables, or use scripts/configure.py:

• HIDREAM_AUTHORIZATION: Bearer token value only
• HIDREAM_ENDPOINT: API Endpoint (default: https://vivago.ai)
• OPENCLAW_AUTHORIZATION (Legacy): Alternative for HIDREAM_AUTHORIZATION
• OPENCLAW_ENDPOINT (Legacy): Alternative for HIDREAM_ENDPOINT

Dependencies

• requests (Python library) - see requirements.txt

Video Model Scripts

• scripts/sora_2_pro.py
• scripts/seedance_1_0_pro.py
• scripts/seedance_1_5_pro.py
• scripts/minimax_hailuo_02.py
• scripts/kling.py (Refactored for Python access)

Image Model Scripts

• scripts/seedream.py (Refactored for Python access)
• scripts/nano_banana.py

Notes

• Per-model scripts only handle parameter parsing and payload building.
• Base scripts handle request submission, polling, and auth.