image-generator

name	image-generator
description	Generate images for WeChat articles using EasyClaw Seedream 5.0 Lite. This skill provides the image-generation primitive used after the agent decides whether inline images are needed. Called by article-writer after article completion. Supports multiple sizes and styles for different use cases.

Image Generator

Generate inline images for WeChat Official Account articles. Uses EasyClaw Seedream 5.0 Lite for AI generation. Authentication is automatically obtained from the OpenClaw runtime (~/.openclaw/), no additional API key configuration is needed when that runtime contains EasyClaw-compatible image settings. If AI is unavailable, the script falls back to a local info-card image.

This skill does not decide by itself whether an article should have images. The agent must make that decision first, then call generate_image.py for each approved image.

Use Cases

• Triggered by article-writer after it decides inline images are needed
• User explicitly requests "add images to this article"
• Generate specific image with custom prompt

Script Directory

This skill's scripts are located in ${SKILL_DIR}/scripts/, where SKILL_DIR is the directory containing this SKILL.md file.

Script	Purpose	Usage
scripts/generate_image.py	Generate single image	python3 ${SKILL_DIR}/scripts/generate_image.py --prompt "描述" -o output.jpg

Configuration

Authentication is automatically obtained from the OpenClaw runtime (~/.openclaw/). No additional API key configuration is needed when that runtime contains EasyClaw-compatible image settings.

Dependencies

pip install openai Pillow

Image Sizes

Size	Use Case	Aspect Ratio
800*600	Default inline image	4:3 landscape
600*800	Portrait image	3:4 portrait
800*800	Square image	1:1 square
1280*720	Cover image (delegates to cover-generator)	16:9 landscape

Note: --size is specified in 宽*高 format but internally converted to an aspect ratio for the Seedream 5.0 Lite model.

Workflow

Mode 1: Agent-Orchestrated Inline Images (Recommended)

After article writing, the agent must first make an image decision, then use this skill to generate the approved images.

Decision protocol:

1. Analyze article theme, emotional tone, and structure
2. Decide image count and positions based on content needs
3. Choose one of two outcomes:
   • Generate images and insert them into Markdown
   • Skip images and explicitly state the reason
4. If generating:
   • Create the images directory
   • Call generate_image.py once per image
   • Insert ![](./images/img_001.jpg) into Markdown
5. If skipping:
   • Record the exact reason in the work summary

Default guidance:

• News brief / very short article: 0-1 image
• Standard article: usually 1-3 images
• Deep analysis article: usually 2-5 images
• For non-news articles, default to at least 1 image unless there is a clear skip reason

Account-specific default for wechat-lobster:

• For promotional articles about 盯钉喵-一人公司智能搭子, prefer 2-3 case-card style images
• Recommended card order:
  1. 流程替代卡
  2. 结果收益卡
  3. 适用人群卡
• Prefer simple layouts, white-background cards, clean icons, and strong result labels
• Skip purely decorative images that do not improve the reader's understanding of value

Mode 2: Single Image Generation

Generate a specific image with custom prompt:

# Basic usage
python3 ${SKILL_DIR}/scripts/generate_image.py --prompt "AI assistant working on laptop" -o image.jpg

# With custom size
python3 ${SKILL_DIR}/scripts/generate_image.py --prompt "Data visualization chart" --size 800*800 -o chart.jpg

# With style hint
python3 ${SKILL_DIR}/scripts/generate_image.py --prompt "Tech startup office" --style tech -o office.jpg

# Skip generation
python3 ${SKILL_DIR}/scripts/generate_image.py --prompt "anything" --no-ai -o image.jpg

Parameters

generate_image.py

Parameter	Description	Default
--prompt	Image description (required for AI mode)	-
--size	Image dimensions (width*height), internally converted to aspect ratio	800*600
--style	Style hint: modern, tech, minimal, warm, bold	modern
-o, --output	Output path	output/images/img_timestamp.jpg
--no-ai	Skip image generation and do not output a file	false
--context	Article context for better prompt generation	-

Style Presets

When --style is specified, prompt is enhanced with style modifiers:

Style	Enhancement
modern	Clean, minimalist, professional lighting
tech	Futuristic, blue/purple tones, digital elements
minimal	Simple shapes, plenty whitespace, muted colors
warm	Warm lighting, cozy atmosphere, soft colors
bold	High contrast, vibrant colors, dramatic

Example:

--prompt "Team meeting" --style warm
→ Actual prompt: "Team meeting, warm lighting cozy atmosphere soft colors"

Prompt Enhancement

For auto-insert mode, prompts are automatically enhanced based on context:

Context Type	Base Prompt	Enhancement
Data/Stats	Abstract data visualization	Modern infographic style, clean charts
Tutorial/How-to	Step-by-step illustration	Minimal UI style, clear icons
Story/Case	Scene description	Cinematic lighting, professional composition
Analysis	Conceptual diagram	Tech style, abstract elements
News	Relevant scene	Photojournalism style, realistic

For wechat-lobster, prefer prompts that translate to:

• workflow replacement
• value summary cards
• audience-specific benefit cards

Output

generate_image.py Output

[AI] 调用 EasyClaw Seedream 5.0 Lite...
[AI] Prompt: ...
[AI] Aspect ratio: 4:3
[OK] 图片已保存：/path/to/image.jpg

Example Agent-Orchestrated Output

[决策] 文章评估完成，计划插入 3 张插图
[位置] 第 2 段后 (字符 456)
[位置] 第 4 段后 (字符 1234)
[位置] 第 6 段后 (字符 2100)

[生成] 图片 1/3：AI工具界面示意图
[OK] 保存至 images/img_001.jpg
[插入] ![](./images/img_001.jpg)

[生成] 图片 2/3：数据可视化图表
[OK] 保存至 images/img_002.jpg
[插入] ![](./images/img_002.jpg)

[生成] 图片 3/3：团队协作场景
[OK] 保存至 images/img_003.jpg
[插入] ![](./images/img_003.jpg)

[完成] 插图决策已执行，共插入 3 张插图

Skip / Failure Behavior

When EasyClaw Is Not Configured

[AI] EasyClaw 配置不可用：Missing required file: ...
[Error] 图片生成失败

The script falls back to a local info-card image so the workflow can continue for preview and publishing prep.

When --no-ai Is Specified

[Skip] 已指定 --no-ai，跳过插图生成

The script exits with code 0 and does not create an image file. The caller should treat this as a recorded skip, not as a successful image insertion.

When API Fails

[AI] API 调用失败：timeout
[Error] 图片生成失败

The script falls back to a local info-card image so the workflow can continue.

Integration with Other Skills

• article-writer: Makes image decisions in Step 8, then calls generate_image.py for each approved image
• cover-generator: Delegates cover generation (1280*720 size)
• publish-orchestrator: Uploads inline images to WeChat and replaces local paths with media URLs

Example Usage in article-writer

After article completion and image decision:

# Step 8: Decide images first
echo "[Step 8] 执行插图决策..."

# Create images directory
mkdir -p "$(dirname "$ARTICLE_PATH")/images"

# Generate one approved image
python3 ${SKILL_DIR}/scripts/generate_image.py \
  --prompt "YOUR DETAILED PROMPT HERE" \
  --style "modern" \
  --size "800*600" \
  -o "$(dirname "$ARTICLE_PATH")/images/img_001.jpg"

# Output
echo "[完成] 插图决策已执行；如生成成功则已插入图片"

Troubleshooting

Issue	Cause	Solution
EasyClaw 配置不可用	OpenClaw runtime lacks EasyClaw-compatible image config	Check ~/.openclaw/ runtime files or set OPENCLAW_EASYCLAW_BASE_URL; local fallback remains available
openai not found	Package not installed	pip install openai
No output image generated	--no-ai specified or EasyClaw unavailable	Check logs to confirm whether generation was intentionally skipped
API 调用失败	Network or API error	Check network connectivity and EasyClaw status
Image too dark/bright	Prompt issue	Add lighting hints to prompt
Image irrelevant	Prompt too vague	Add more context via --context