AIwork4me/ernie-image-skill

ernie-image-skill is a Claude Code skill for generating local PNG images with Baidu AI Studio's ERNIE-Image and ERNIE-Image-Turbo models through an OpenAI-compatible API.

It is designed for explicit ERNIE/Baidu image-generation requests, especially Chinese-language text-to-image workflows. Prompts are sent to Baidu AI Studio, so do not include secrets, credentials, private personal data, or confidential business content.

Features

• ERNIE-Image-Turbo for fast drafts and ERNIE-Image for higher-quality outputs.
• Seven supported sizes across square, landscape, and portrait ratios.
• Batch generation from 1 to 4 images.
• Optional seed for reproducible generations.
• Adjustable inference steps, guidance scale, and prompt enhancement.
• Local PNG output with MEDIA:<absolute-path> lines for compatible clients.
• JSON output for automation and tests.
• Safe output filenames that avoid path traversal and accidental overwrites.

Install as a Claude Code Skill

Copy this repository into a skill directory named ernie-image-skill:

mkdir -p ~/.claude/skills
cp -r ernie-image-skill ~/.claude/skills/ernie-image-skill

Project-level install:

mkdir -p .claude/skills
cp -r ernie-image-skill .claude/skills/ernie-image-skill

The directory name must match the skill name in SKILL.md: ernie-image-skill.

Prerequisites

• Python 3.11+
• uv
• Baidu AI Studio access token: https://aistudio.baidu.com/account/accessToken

Set your API key locally:

export AI_STUDIO_API_KEY="your_access_token"

PowerShell:

$env:AI_STUDIO_API_KEY = "your_access_token"

CLI Examples

Generate one image:

uv run scripts/generate.py "A cinematic sunset over the ocean, wide angle, warm light"

Use the higher-quality model:

uv run scripts/generate.py "A futuristic city at night, rain, neon reflections" --model ERNIE-Image --size 1376x768 --steps 16 --guidance 3.0

Chinese prompt:

uv run scripts/generate.py "一只可爱的橘猫坐在窗台上，柔和晨光，摄影风格" --size 1024x1024

Structured JSON output:

uv run scripts/generate.py "minimal product photo of a ceramic mug" --json

CLI Options

Option	Default	Description
prompt	required	Image description, max 1024 characters
--model	ERNIE-Image-Turbo	ERNIE-Image or ERNIE-Image-Turbo
--size	1024x1024	Image dimensions
--n	1	Number of images, 1-4
--output	.	Output directory
--format	b64_json	API response format: b64_json or url
--seed	random	Reproducibility seed
--steps	provider default	Inference steps, 4-20
--guidance	provider default	Guidance scale, 1.0-7.5
--use-pe	off	Enable prompt enhancement
--prefix	ernie	Safe output filename prefix
--json	off	Print structured JSON

Supported Sizes

Size	Ratio	Best for
1024x1024	1:1	General purpose
1376x768	16:9	Headers, wallpapers, presentation visuals
1264x848	3:2	Photography-style landscapes
1200x896	4:3	Product shots and traditional landscape layouts
896x1200	3:4	Portrait cards and print layouts
848x1264	2:3	Posters, covers, portraits
768x1376	9:16	Mobile wallpapers and story covers

Development

Run tests:

uv run python -m unittest discover -s tests

Validate the skill structure from a directory named ernie-image-skill:

uvx --from skills-ref agentskills validate /path/to/ernie-image-skill

License

MIT-0