ERNIE-Image Skill

Generate images with Baidu AI Studio's ERNIE-Image models through the OpenAI-compatible API. This skill is best when the user specifically wants Baidu/ERNIE generation, Chinese-language image prompts, or local PNG outputs from a configured AI_STUDIO_API_KEY.

Safety Boundary

• Use this skill only for explicit image-generation requests.
• Prompts and generation parameters are sent to Baidu AI Studio. Do not include secrets, credentials, private personal data, or confidential business data in prompts.
• Follow Baidu AI Studio terms and applicable platform rules. Decline unsafe, deceptive, exploitative, or rights-violating image requests.
• Do not print, store, or ask the user to paste AI_STUDIO_API_KEY into chat. The script reads it from the environment only.
• Generated files are written locally. Confirm the output directory when the user has not specified one.

Models

Chinese prompts usually work especially well with these models.

Before Running

Confirm that AI_STUDIO_API_KEY is set in the shell environment. If it is missing, tell the user to create an access token at https://aistudio.baidu.com/account/accessToken and set it locally.

Prefer ERNIE-Image-Turbo, 1024x1024, n=1, and b64_json unless the user asks for different parameters.

Execution

Run from the skill directory so uv can use the bundled project metadata and lockfile:

cd "${CLAUDE_SKILL_DIR}" && uv run scripts/generate.py "<PROMPT>" --model ERNIE-Image-Turbo --size 1024x1024

Batch generation:

cd "${CLAUDE_SKILL_DIR}" && uv run scripts/generate.py "<PROMPT>" --n 4 --output "<output_dir>"

Higher-quality generation:

cd "${CLAUDE_SKILL_DIR}" && uv run scripts/generate.py "<PROMPT>" --model ERNIE-Image --steps 16 --guidance 3.5

Reproducible generation:

cd "${CLAUDE_SKILL_DIR}" && uv run scripts/generate.py "<PROMPT>" --seed 42

Structured output:

cd "${CLAUDE_SKILL_DIR}" && uv run scripts/generate.py "<PROMPT>" --json

Parameters

Select vertical sizes for posters, portraits, and mobile layouts; horizontal sizes for covers, wallpapers, and presentation visuals; square for general use.

Output

The script saves PNG files and prints one MEDIA:<absolute-path> line per image for compatible clients. With --json, it prints:

{
  "success": true,
  "model": "ERNIE-Image-Turbo",
  "prompt": "...",
  "parameters": {"size": "1024x1024", "n": 1, "seed": 42},
  "files": [{"path": "/abs/path/ernie_20260502_140000.png", "size_bytes": 1715660}]
}

The script avoids overwriting existing files by adding a numeric suffix when needed.

Useful Triggers

Treat the text after these explicit triggers as the prompt:

• English: ernie image: ..., baidu image: ..., AI Studio image: ...
• Chinese: 文心生图：..., 百度生图：..., 生成图片：...

For full API notes, prompt guidance, and troubleshooting, read references/api-guide.md.