Communitygithub.com

Agents365-ai/imagenCN

AI image generation via Alibaba Bailian, ByteDance Volcano Ark, and Tencent Hunyuan — 23 models, Chinese text excellence

imagenCN 是什麼？

imagenCN is a Claude Code agent skill that aI image generation via Alibaba Bailian, ByteDance Volcano Ark, and Tencent Hunyuan — 23 models, Chinese text excellence.

相容平台✓Claude Code✓Codex CLI~Cursor✓Gemini CLI
npx skills add Agents365-ai/imagenCN
查看原文→瀏覽所有技能

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

OpenAIChatGPTClaudeClaudeGeminiGeminiGrokGrokPerplexityPerplexityDeepSeekDeepSeek

說明文件

ImagenCN - Alibaba Cloud Bailian Text-to-Image Skill

Overview

Generate images using Alibaba Cloud Bailian API. Default endpoint is China region.

Supports five platforms across nine model families:

  • Alibaba Cloud Bailian (DashScope): Qwen-Image 2.0, Qwen-Image Edit, Qwen-Image legacy, Wan Series, Z-Image
  • ByteDance Volcano Ark: Doubao-Seedream series (OpenAI-compatible REST)
  • Tencent Hunyuan: Hunyuan Image 3.0 (OpenAI-compatible REST)
  • Zhipu / BigModel: CogView-4 and GLM-Image (OpenAI-compatible REST)
  • StepFun / 阶跃星辰: Step-2X and Step-Image-Edit (OpenAI-compatible REST)

Cross-platform support: Windows, macOS, Linux

When to Use This Skill

Automatically activate this skill when:

  • User requests image generation with Chinese text or calligraphy
  • Need photorealistic images or photography-style visuals
  • Creating commercial posters, illustrations, or digital art
  • User mentions any of these: Alibaba Cloud / Bailian / Qwen / Wan / DashScope, ByteDance / Volcano Ark / Seedream / Doubao, Tencent / Hunyuan
  • Any task where AI-generated image with strong Chinese support would be helpful

Workflow

Step 1 — Refine the prompt (interactive, never skip)

Users often give short, casual descriptions ("生成一只猫"). Before calling the API, present 3 refined prompt options with different style directions. Add, as appropriate:

  • Subject details (shape, colour, material, expression, pose)
  • Lighting (golden hour, studio, rim light, soft diffused, neon, cinematic)
  • Composition (rule of thirds, shallow depth of field, wide shot, close-up)
  • Style / medium (photorealistic, oil painting, watercolour, 3D render, vector)
  • Mood / atmosphere (serene, dramatic, whimsical, dystopian, elegant)
  • Quality keywords (8K, hyperdetailed, award-winning, professional photography)
  • For Chinese text on images: text content, placement, font style, colour, size

Label the options clearly (e.g. A / B / C) with a one-line summary of each direction. Let the user pick one, combine elements from multiple, or request a new direction. Iterate until they confirm ("go", "generate", "ok", etc.), then proceed to generation.

Step 2 — Pick a model

Choose based on the request (see Model Selection Guide below). Default to qwen-image-2.0-pro if unsure. Mention your choice to the user.

Step 3 — Pick a size

Native 2K for Qwen-Image 2.0, 1K/2K/4K for Wan2.7, or an aspect-ratio preset (16:9, 1:1, etc.).

Step 4 — Generate

Run scripts/generate_image.py with the confirmed prompt and output path.

Step 5 — Save

If the output path was implicit, save into the user's current working directory.

Models

Qwen-Image 2.0 family - Latest Flagship (MultiModalConversation API)

ModelDescription
qwen-image-2.0-proDefault. Latest flagship, native 2K, strongest typography and detail
qwen-image-2.0-pro-2026-06-22Latest snapshot (Jun 2026): generation + editing fusion, better text rendering and prompt adherence
qwen-image-2.0Standard 2.0 tier, native 2K
qwen-image-maxPrevious-gen flagship (Dec 2025)
qwen-image-max-2025-12-30qwen-image-max snapshot: improved realism, fewer AI artifacts

Qwen-Image Edit family - Image Editing (MultiModalConversation API)

Editing models require an input image via --image (local path or URL). Omit --size to match the input image dimensions.

ModelDescription
qwen-image-edit-maxFlagship editing model, strongest instruction following
qwen-image-edit-max-2026-01-16Latest max snapshot (Jan 2026)
qwen-image-edit-plusFaster, lower-cost editing

Qwen-Image legacy (ImageSynthesis API)

ModelDescription
qwen-image-plusDistilled accelerated version of qwen-image-max
qwen-image-plus-2026-01-09qwen-image-plus snapshot (Jan 2026): faster high-quality generation
qwen-imageBase model

Wan Series - Photorealistic Generation (ImageGeneration API)

ModelDescription
wan2.7-image-proLatest. Up to 4K output, unified architecture (T2I + edit + multi-image)
wan2.7-imageWan 2.7 standard, up to 2K
wan2.6-t2iWan 2.6, flexible sizing
wan2.5-t2i-previewHigh quality, up to 768x2700
wan2.2-t2i-flashSpeed-optimized
wan2.2-t2i-plusProfessional tier
wanx2.1-t2i-turboFast execution
wanx2.1-t2i-plusProfessional tier
wanx2.0-t2i-turboEarlier generation

Z-Image - Lightweight & Fast (MultiModalConversation API)

ModelDescription
z-image-turboFast, low-cost generation; bilingual (CN/EN) text rendering, high-fidelity portraits and product images. Pixel area 512x512 to 2048x2048

Volcano Ark - ByteDance Seedream (OpenAI-compatible API)

ModelDescription
doubao-seedream-5-0-260128Ark default. Latest, up to 3K, PNG/JPEG output, best text rendering
doubao-seedream-4-5-251128Seedream 4.5, up to 4K
doubao-seedream-4-0-250828Seedream 4.0, up to 4K, budget-friendly

Tencent Hunyuan (OpenAI-compatible API)

ModelDescription
hy-image-v3.0Hunyuan default. Flagship 3.0, strong composition awareness, handles complex Chinese prompts up to 8K chars

Zhipu / BigModel - CogView-4 & GLM-Image (OpenAI-compatible API)

ModelDescription
cogview-4Zhipu default. Stable alias for latest CogView-4, native Chinese text rendering
cogview-4-250304CogView-4 fixed snapshot (Mar 2025), reproducible results
glm-imageGLM-Image flagship, up to 2048x2048, hybrid autoregressive/diffusion

StepFun / 阶跃星辰 - Step-2X (OpenAI-compatible API)

ModelDescription
step-2x-largeStepFun default. High quality (0.1 RMB/image), up to 1024x1024
step-image-edit-2Fast & cheap (0.02 RMB/image), supports negative prompts, 8 inference steps

Usage

Basic Usage

# Default model (qwen-image-2.0-pro, native 2K output)
python ~/.claude/skills/imagenCN/scripts/generate_image.py "A cute cat" output.png

# Photorealistic with Wan model (Wan2.7 supports 4K)
python ~/.claude/skills/imagenCN/scripts/generate_image.py --model wan2.7-image-pro --size 4K "Realistic photo of mountains at sunset" photo.png

# Edit an existing image (requires --image; local path or URL)
python ~/.claude/skills/imagenCN/scripts/generate_image.py --model qwen-image-edit-max --image input.png "Change the background to a beach at sunset" edited.png

Size Options

# Use ratio preset
python ~/.claude/skills/imagenCN/scripts/generate_image.py --size 16:9 "Wide landscape" landscape.png

# Use exact dimensions
python ~/.claude/skills/imagenCN/scripts/generate_image.py --size 1280*720 "Custom size" custom.png

Size Presets

Qwen-Image 2.0 (native 2K):

  • 1:1 -> 2048x2048 (default)
  • 16:9 -> 2688x1536
  • 9:16 -> 1536x2688
  • 4:3 -> 2304x1728
  • 3:4 -> 1728x2304
  • 1K -> 1024x1024
  • 2K -> 2048x2048

Qwen-Image legacy:

  • 1:1 -> 1328x1328
  • 16:9 -> 1664x928
  • 9:16 -> 928x1664
  • 4:3 -> 1472x1104
  • 3:4 -> 1104x1472

Z-Image (pixel area 512x512 to 2048x2048):

  • 1:1 -> 1024x1024 (default)
  • 16:9 -> 1280x720
  • 9:16 -> 720x1280
  • 2:3 -> 1024x1536
  • 3:2 -> 1536x1024
  • 1K -> 1024x1024

Wan Series (Wan2.7 also accepts 1K/2K/4K):

  • 1:1 -> 1024x1024
  • 1:1-large -> 1280x1280
  • 16:9 -> 1280x720
  • 9:16 -> 720x1280
  • 4:3 -> 1200x900
  • 3:4 -> 900x1200
  • 2:1 -> 1440x720

Volcano Ark (Seedream):

  • 1:1 -> 2048x2048
  • 16:9 -> 2848x1600
  • 9:16 -> 1600x2848
  • 4:3 -> 2304x1728
  • 3:4 -> 1728x2304
  • 3:2 -> 2496x1664
  • 2:3 -> 1664x2496
  • 1K / 2K / 3K / 4K (model-dependent max resolution)

Tencent Hunyuan (colon-separated format):

  • 1:1 -> 1024:1024
  • 16:9 -> 1920:1080
  • 9:16 -> 1080:1920
  • 4:3 -> 1600:1200
  • 3:4 -> 1200:1600

Zhipu (CogView-4 / GLM-Image):

  • 1:1 -> 1024x1024 (default)
  • 16:9 -> 1344x768
  • 9:16 -> 768x1344
  • 4:3 -> 1152x864
  • 3:4 -> 864x1152
  • 2:1 -> 1440x720
  • 1:2 -> 720x1440

StepFun (Step-2X):

  • 1:1 -> 1024x1024 (default)
  • 1:1-small -> 512x512
  • 16:9 -> 1280x800
  • 9:16 -> 800x1280

Advanced Options

# With negative prompt
python ~/.claude/skills/imagenCN/scripts/generate_image.py --negative "blurry, low quality" "High quality portrait" portrait.png

# List all models
python ~/.claude/skills/imagenCN/scripts/generate_image.py --list-models

Requirements

pip install dashscope requests

# Optional: for coloured output and styled tables
pip install rich

Environment Variables

# Alibaba Cloud Bailian (DashScope)
export DASHSCOPE_API_KEY="your_api_key"        # Required
export DASHSCOPE_MODEL="wan2.7-image-pro"       # Optional default model
export DASHSCOPE_API_BASE="cn"                  # Optional: cn, sg, us

# ByteDance Volcano Ark
export ARK_API_KEY="your_api_key"               # Required for Ark
export ARK_MODEL="doubao-seedream-5-0-260128"   # Optional default model

# Tencent Hunyuan (TokenHub)
export HUNYUAN_API_KEY="your_api_key"           # Required for Hunyuan
export HUNYUAN_MODEL="hy-image-v3.0"            # Optional default model

# Zhipu / BigModel
export ZHIPUAI_API_KEY="your_api_key"           # Required for Zhipu
export ZHIPUAI_MODEL="cogview-4"                # Optional default model

# StepFun / 阶跃星辰
export STEP_API_KEY="your_api_key"              # Required for StepFun
export STEP_MODEL="step-2x-large"               # Optional default model

Get API Keys:

Config File (Optional)

Create ~/.imagenCN.json for personal defaults, or .imagenCN.json in a project directory for per-project overrides. API keys stay in environment variables for security.

{
  "platform": "ark",
  "model": "doubao-seedream-5-0-260128",
  "size": "2K"
}

All keys are optional. Priority (highest first):

  1. CLI arguments (--platform, --model, --size)
  2. Project config (.imagenCN.json in current directory)
  3. User config (~/.imagenCN.json)
  4. Environment variables (DASHSCOPE_MODEL, ARK_MODEL, HUNYUAN_MODEL)
  5. Built-in defaults

API Endpoints

RegionAliasURL
China (default)cnhttps://dashscope.aliyuncs.com/api/v1
Singaporesghttps://dashscope-intl.aliyuncs.com/api/v1
Virginiaushttps://dashscope-us.aliyuncs.com/api/v1
# Switch to Singapore endpoint
export DASHSCOPE_API_BASE="sg"

# Or use full URL
export DASHSCOPE_API_BASE="https://dashscope-intl.aliyuncs.com/api/v1"

Model Selection Guide

Quick Pick — You Only Need Eight

What you wantModelPlatform
Default / general (posters, text)qwen-image-2.0-proDashScope
Photorealistic (portraits, landscapes)wan2.7-image-proDashScope
Edit an imageqwen-image-edit-maxDashScope
Cheap & fastz-image-turboDashScope
Photo + text combodoubao-seedream-5-0-260128Volcano Ark
Complex Chinese compositionhy-image-v3.0Tencent Hunyuan
Chinese text in imagescogview-4Zhipu
Ultra-cheap volume genstep-image-edit-2StepFun

All other models are legacy/snapshot variants.

Full Reference

Use CaseRecommended Model
General high-quality (default)qwen-image-2.0-pro
Chinese text/calligraphyqwen-image-2.0-pro
English text on imagesqwen-image-2.0-pro
Posters with typographyqwen-image-2.0-pro
Photorealistic photos (4K)wan2.7-image-pro
Photorealistic photos (2K)wan2.7-image
Portrait photographywan2.7-image-pro
Image editing (best quality)qwen-image-edit-max
Image editing (fast, low-cost)qwen-image-edit-plus
Fast, low-cost generationz-image-turbo
High-fidelity portraits / product shots (fast)z-image-turbo
Fast photorealistic (Wan)wan2.2-t2i-flash
Lower-cost text renderingqwen-image-plus
ByteDance best qualitydoubao-seedream-5-0-260128
Budget-friendly 4K (ByteDance)doubao-seedream-4-0-250828
Complex Chinese prompts (Tencent)hy-image-v3.0

Platform Quick Comparison

FeatureDashScopeArkHunyuanZhipuStepFun
Best forText, varietyPhoto+textComplex CNCN text in imageUltra-cheap
Max res4K4K2K2K1K
SDKdashscopeNoneNoneNoneNone
PriceVaries~0.22~0.20~0.06~0.02
Env varDASHSCOPE_API_KEYARK_API_KEYHUNYUAN_API_KEYZHIPUAI_API_KEYSTEP_API_KEY

Comparison with Imagen (Gemini)

FeatureImagenCN (Bailian)Imagen (Gemini)
Chinese text renderingExcellentGood
English text renderingExcellentGood
Photorealistic imagesExcellentGood
SpeedMediumFast
Model variety15+ models3 models
Max resolution4K (Wan2.7-Pro)2K

Examples

Volcano Ark (ByteDance)

# Default Ark model (Seedream 5.0)
ARK_API_KEY="xxx" python scripts/generate_image.py \
  --platform ark \
  "A vibrant close-up editorial portrait, Vogue magazine cover style" \
  portrait.png

# With 4K output
ARK_API_KEY="xxx" python scripts/generate_image.py \
  --platform ark --model doubao-seedream-4-5-251128 --size 4K \
  "Breathtaking mountain sunset, golden hour, professional photography" \
  landscape.png

Tencent Hunyuan

# Default Hunyuan model (Image 3.0)
HUNYUAN_API_KEY="xxx" python scripts/generate_image.py \
  --platform hunyuan \
  "An astronaut riding a horse on the moon, cinematic lighting, 8K detail" \
  scifi.png

# With prompt auto-enhance disabled
HUNYUAN_API_KEY="xxx" python scripts/generate_image.py \
  --platform hunyuan --revise 0 \
  "A cute orange cat napping in sunlight, oil painting style" \
  cat.png

Chinese New Year Poster (DashScope)

python ~/.claude/skills/imagenCN/scripts/generate_image.py \
  "A beautiful Chinese New Year poster with red background, golden text, fireworks and firecrackers" \
  new_year_poster.png

Photorealistic Landscape (4K)

python ~/.claude/skills/imagenCN/scripts/generate_image.py \
  --model wan2.7-image-pro \
  --size 4K \
  "Breathtaking sunset over mountain range, golden hour, professional photography" \
  landscape.png

Product Shot

python ~/.claude/skills/imagenCN/scripts/generate_image.py \
  --model wan2.7-image \
  --size 2K \
  "Professional product photography of a coffee cup on marble surface, studio lighting" \
  product.png

相關技能

steipete/sag

ElevenLabs text-to-speech with mac-style say UX.

community

steipete/oracle

Oracle CLI second-model review/debug/refactor/design with selected files, dry-run token checks, API or browser engine.

community

steipete/peekaboo

Capture and automate macOS UI with the Peekaboo CLI.

community

obra/brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

community

affaan-m/prisma-patterns

Prisma ORM patterns for TypeScript backends — schema design, query optimization, transactions, pagination, and critical traps like updateMany returning count not records, $transaction timeouts, migrate dev resetting the DB, @updatedAt skipped on bulk writes, and serverless connection exhaustion.

community

affaan-m/django-celery

Django + Celery async task patterns — configuration, task design, beat scheduling, retries, canvas workflows, monitoring, and testing. Use when adding background jobs, scheduled tasks, or async processing to a Django app.

community