panda-imagine

多 Provider 图片生成引擎。支持 OpenAI、Azure、Google Gemini、OpenRouter、 通义万相(DashScope)、MiniMax、即梦(Jimeng)、豆包(Seedream)、Replicate。 支持文生图、参考图、宽高比、批量生成。 触发词："生成图片"、"画图"、"g...

Security Scan

The skill's code, runtime instructions, and requested configuration are consistent with a multi-provider image-generation CLI; nothing indicates hidden or unrelated capabilities.

Purpose & Capability

Name/description claim a multi-provider image generator and the code + SKILL.md implement connectors for OpenAI, Google, Azure, OpenRouter, DashScope, MiniMax, Jimeng, Seedream and Replicate; required binaries (bun or npx) and listed environment variables directly map to those providers.

Instruction Scope

SKILL.md and scripts read/save an EXTEND.md config, load .panda-skills/.env and ~/.panda-skills/.env (and .baoyu-skills fallbacks), accept user-supplied prompt files and reference images, and (for Jimeng) detect/run a local CLI. These actions are expected for a CLI which needs provider keys and per-project prefs, but they do access files in the user's home/project config locations and will load environment variables from disk (which can contain secrets).

Install Mechanism

No formal install spec is provided (instruction-only), but runtime instructions execute the included TypeScript with bun or via npx -y bun. Using npx to fetch/run bun and the SKILL.md suggestion to install the jimeng 'dreamina' CLI via curl | bash are standard but inherently riskier than purely local code because they fetch remote tooling at runtime—optional but worth reviewing.

Credentials

The environment variables enumerated (OPENAI_API_KEY, GOOGLE_API_KEY, AZURE_OPENAI_API_KEY, REPLICATE_API_TOKEN, JIMENG_* etc.) align with the declared provider integrations. The skill will load env values from multiple .env locations and EXTEND.md which is reasonable for pref/config override behavior.

Persistence & Privilege

The skill does not request always:true or other elevated persistence. It may write its own EXTEND.md preferences into .panda-skills or the user-chosen path, which is normal for a CLI-style skill.

Assessment

This skill appears to do what it says: unify multiple image-generation providers. Before installing, review and accept these points: (1) It will load/save preferences (EXTEND.md) and read .panda-skills/.env and ~/.panda-skills/.env (so API keys or other env values stored there can be read). (2) Runtime runs the included TypeScript via bun or npx -y bun—npx may fetch packages on demand. (3) For the Jimeng provider the skill can detect and run a local 'dreamina' CLI and suggests installing it via curl | bash (optional but be cautious fetching remote install scripts). (4) Only provide API keys with least privilege to providers you trust and avoid running the skill in environments containing unrelated high-value credentials. If you want stronger assurance, review the included scripts (providers/*.ts and main.ts) yourself or run the tool in an isolated environment/container.

图片生成引擎

多 Provider 图片生成引擎，通过统一的 CLI 接口调用 9 种图片生成服务。

脚本路径

1. {baseDir} = 本 SKILL.md 所在目录
2. 脚本路径 = {baseDir}/scripts/main.ts
3. 运行时: bun 可用 → bun; 否则 npx -y bun

步骤 0: 加载偏好 ⛔ 阻塞

检查 EXTEND.md：

test -f .panda-skills/panda-imagine/EXTEND.md && echo "project"
test -f "$HOME/.panda-skills/panda-imagine/EXTEND.md" && echo "user"

结果	操作
找到	加载、解析、应用
未找到	⛔ 运行 首次配置 → 保存 → 继续

兼容读取: .baoyu-skills/baoyu-imagine/EXTEND.md（fallback）

使用方式

# 基本用法
${BUN_X} {baseDir}/scripts/main.ts --prompt "一只猫" --image cat.png

# 指定宽高比
${BUN_X} {baseDir}/scripts/main.ts --prompt "风景画" --image out.png --ar 16:9

# 高质量
${BUN_X} {baseDir}/scripts/main.ts --prompt "一只猫" --image out.png --quality 2k

# 从提示词文件生成
${BUN_X} {baseDir}/scripts/main.ts --promptfiles prompt.md --image out.png

# 参考图（Google, OpenAI, Azure, OpenRouter, Replicate, MiniMax, Seedream）
${BUN_X} {baseDir}/scripts/main.ts --prompt "改为蓝色" --image out.png --ref source.png

# 通义万相
${BUN_X} {baseDir}/scripts/main.ts --prompt "一只可爱的猫" --image out.png --provider dashscope

# MiniMax（角色参考图）
${BUN_X} {baseDir}/scripts/main.ts --prompt "站在窗前" --image out.jpg --provider minimax --ref portrait.png

# 即梦（自动检测 dreamina CLI 或 API 模式）
${BUN_X} {baseDir}/scripts/main.ts --prompt "科技海报" --image out.png --provider jimeng

# 即梦 + 参考图（需要 dreamina CLI）
${BUN_X} {baseDir}/scripts/main.ts --prompt "改为水彩风格" --image out.png --provider jimeng --ref source.png

# 豆包 Seedream
${BUN_X} {baseDir}/scripts/main.ts --prompt "一只猫" --image out.png --provider seedream

# 批量模式
${BUN_X} {baseDir}/scripts/main.ts --batchfile batch.json

批量文件格式

{
  "jobs": 4,
  "tasks": [
    {
      "id": "hero",
      "promptFiles": ["prompts/hero.md"],
      "image": "out/hero.png",
      "ar": "16:9"
    }
  ]
}

选项

选项	描述
--prompt <text>, -p	提示词文本
--promptfiles <files...>	从文件读取提示词（多文件拼接）
--image <path>	输出图片路径
--batchfile <path>	批量任务 JSON 文件
--jobs <count>	批量 worker 数量（默认自动，上限见配置）
--provider <name>	强制指定 Provider
--model <id>, -m	模型 ID
--ar <ratio>	宽高比（如 16:9, 3:4）
--size <WxH>	尺寸（如 1024x1024）
--quality normal|2k	质量预设（默认 2k）
--imageSize 1K|2K|4K	图片尺寸（Google/OpenRouter）
--ref <files...>	参考图
--n <count>	生成数量
--json	JSON 输出

环境变量

变量	描述
OPENAI_API_KEY	OpenAI API 密钥
AZURE_OPENAI_API_KEY	Azure OpenAI API 密钥
AZURE_OPENAI_BASE_URL	Azure 资源端点
GOOGLE_API_KEY	Google API 密钥
OPENROUTER_API_KEY	OpenRouter API 密钥
DASHSCOPE_API_KEY	通义万相 API 密钥
MINIMAX_API_KEY	MiniMax API 密钥
REPLICATE_API_TOKEN	Replicate API 令牌
JIMENG_ACCESS_KEY_ID	即梦火山引擎 Access Key（API 模式）
JIMENG_SECRET_ACCESS_KEY	即梦火山引擎 Secret Key（API 模式）
ARK_API_KEY	豆包 Seedream ARK API 密钥

加载优先级: CLI 参数 > EXTEND.md > 环境变量 > .panda-skills/.env > ~/.panda-skills/.env

即梦 Provider 无需设置环境变量——安装 dreamina CLI 并登录后即可使用。

Provider 选择

Provider	默认模型	参考图
google	gemini-3-pro-image-preview	✓
openai	gpt-image-1.5	✓
azure	gpt-image-1.5	✓
openrouter	gemini-3.1-flash-image-preview	✓
dashscope	qwen-image-2.0-pro	✗
minimax	image-01	✓
jimeng	jimeng_t2i_v40	✓(CLI)
seedream	doubao-seedream-5-0	✓
replicate	google/nano-banana-pro	✓

即梦双模式:

• CLI 模式（优先）：安装了 dreamina CLI → 自动使用，支持 --ref 参考图
• API 模式（fallback）：设置了 JIMENG_ACCESS_KEY_ID → 使用火山引擎 HMAC 签名，不支持 --ref

安装 dreamina CLI:

curl -fsSL https://jimeng.jianying.com/cli | bash
dreamina login

自动选择逻辑:

1. 有 --ref 且无 --provider → 按 google > openai > azure > openrouter > replicate > seedream > minimax > jimeng 顺序
2. 指定 --provider → 使用该 Provider
3. 仅一个 API Key（或 dreamina CLI）→ 使用对应 Provider
4. 多个可用 → 使用第一个

质量预设

预设	Google	OpenAI	用途
normal	1K	1024px	快速预览
2k（默认）	2K	2048px	封面、插画、信息图

错误处理

• API Key 缺失 → 报错并提示设置
• 生成失败 → 自动重试最多 3 次
• 参考图 + 不支持的 Provider → 报错并提示替代方案

参考文档

文件	内容
references/config/first-time-setup.md	首次配置流程
references/config/preferences-schema.md	EXTEND.md schema