qwen-image-gen

基于阿里云百炼 Qwen-Image 文生图模型的生图 skill。支持同步生成、异步任务轮询、下载生成结果到本地。 当用户需要根据提示词生成图片、批量出图、指定尺寸/比例，或继续查询已有图片生成任务时，使用此 skill。

Security Scan

The skill's requirements, instructions, and included script align with its stated purpose (calling Aliyun DashScope/Qwen-Image APIs to generate and download images); nothing requested appears disproportionate or unrelated.

Purpose & Capability

The skill is an image-generation client for Aliyun's Qwen-Image service. It requires Node and a DASHSCOPE_API_KEY, references dashscope.aliyuncs.com as the API host, and includes code to call the documented endpoints — these are appropriate for the described purpose.

Instruction Scope

SKILL.md instructs the agent to set DASHSCOPE_API_KEY, optionally use config.json, run the provided Node script, and download outputs to outputs/. The instructions only reference files and env vars relevant to configuring and running the image-generation flow (no unrelated file reads or external endpoints).

Install Mechanism

There is no install spec (instruction-only behavior plus an included Node script). The script relies on Node >=18 and built-in fetch; no external download/install from untrusted URLs or package registries occurs. Risk from install mechanism is low.

Credentials

The only required credential is DASHSCOPE_API_KEY (primaryEnv). That matches the documented API usage. The code may also read optional environment keys (e.g., QWEN_IMAGE_MODEL) and a local config.json; these are reasonable and proportional to configuration needs.

Persistence & Privilege

The skill does not request permanent always:true inclusion and does not attempt to modify other skills or system-wide settings. It runs as a local Node script and writes outputs to a local outputs/ directory as expected.

Assessment

This skill runs a local Node script that will call Aliyun's DashScope (Qwen-Image) APIs and download generated PNGs to an outputs/ directory. Before installing/providing credentials: 1) confirm the DASHSCOPE_API_KEY is the correct key for the intended Aliyun region (the README warns region keys/URLs cannot be mixed), 2) review the included scripts (they are present and readable) and run in an isolated environment if unsure, 3) store the API key in an environment variable rather than committing it to a repo or config file, and 4) be aware that generating images incurs cost per image per the pricing table. If you do not trust the source, do not provide high-privilege credentials or run the script on sensitive systems.

通过阿里云百炼 Qwen-Image 接口生成图片。默认使用 qwen-image-2.0-pro，支持按模型自动选择同步/异步模式，并在成功后下载 PNG 到本地。

详细接口、尺寸限制、价格表见 references/api.md。

Setup

1. 配置 API Key：

export DASHSCOPE_API_KEY="sk-xxx"

也可以在 skill 目录下创建 config.json，参考 config.example.json。

2. 运行环境要求 Node.js >= 18

3. 北京和新加坡地域使用不同的 API Key 与 Base URL，不能混用

配置优先级：

• CLI 参数
• 环境变量
• config.json
• 脚本默认值

默认模型

• 默认模型：qwen-image-2.0-pro
• qwen-image-2.0-pro、qwen-image-2.0、qwen-image-max 走同步接口
• qwen-image-plus、qwen-image 走异步接口

支持范围

• qwen-image-2.0-pro / qwen-image-2.0：支持 1-6 张，适合通用文生图
• qwen-image-max：固定 1 张，偏真实感
• qwen-image-plus / qwen-image：固定 1 张，支持异步任务
• 图片输出格式为 png
• 任务和结果链接默认保留 24 小时，拿到后要尽快下载

CLI

主要脚本：node scripts/qwen-image-gen.js

默认出图

node scripts/qwen-image-gen.js \
  --prompt="一间有着精致窗户的花店，漂亮的木质门，摆放着花朵" \
  --goal="quality"

低成本试探

node scripts/qwen-image-gen.js \
  --prompt="一个赛博朋克风格的雨夜街道" \
  --goal="cheap" \
  --n=1

异步任务

node scripts/qwen-image-gen.js \
  --prompt="复古海报风格的城市街景" \
  --model="qwen-image-plus" \
  --mode="async"

继续查询已有任务

node scripts/qwen-image-gen.js \
  --task-id="你的task_id"

仅做预检

node scripts/qwen-image-gen.js \
  --prompt="一张产品海报，主体是一瓶绿色玻璃香水" \
  --ratio="3:4" \
  --dry-run

参数约定

• --prompt：正向提示词，提交新任务时必填
• --negative-prompt：反向提示词，建议控制在 500 字符内
• --goal=cheap|balanced|quality：按用户意图选择默认模型和尺寸
• --tier=draft|standard|final：按档位选择模型
• --model：模型名，默认 qwen-image-2.0-pro
• --mode=auto|sync|async：自动选择或强制接口模式
• --ratio：快捷比例
  • qwen-image-2.0*：1:1 3:4 4:3 9:16 16:9
  • qwen-image-max / qwen-image-plus / qwen-image：1:1 3:4 4:3 9:16 16:9
• --size：具体分辨率，如 2048*2048
• --n：张数
  • qwen-image-2.0*：1-6
  • 其他模型：固定 1
• --prompt-extend=true|false：是否启用提示词改写，默认 true
• --watermark=true|false：是否添加水印，默认 false
• --task-id：查询已有异步任务
• --no-wait：只提交异步任务，不继续轮询
• --poll-interval：轮询间隔秒数，默认 10
• --timeout：总等待超时秒数，默认 600
• --output-dir：图片下载目录，默认 outputs/
• --name：输出文件名前缀
• --dry-run：只打印最终请求体和配置预检

工作流

1. 用户要生图时，优先提交新任务而不是直接假定已有结果。
2. qwen-image-2.0-pro / qwen-image-2.0 / qwen-image-max 直接同步出图并下载。
3. qwen-image-plus / qwen-image 走异步任务，返回 task_id 后轮询。
4. 成功后立即下载生成结果到本地 outputs/。
5. 需要稳定复现时，显式传 --seed。

价格提醒

• 价格按中国内地单价表估算；如果使用新加坡地域，脚本会切换到国际单价表
• 计费按成功生成的图片张数计算
• 免费额度和单价会随官方更新变化，需同步更新脚本内置表