魔搭图片生成

魔搭(ModelScope)AI 图片生成。支持多种模型、LoRA 微调。触发词：生成图片、AI绘画、文生图、image generation、generate image。当用户要求生成图片、画图、AI 作画，或提到魔搭、ModelScope时使用。

Security Scan

The skill appears to implement the stated ModelScope image-generation functionality, but its metadata omits the required API key and a few small inconsistencies reduce confidence.

Purpose & Capability

The skill's name/description and code match: it calls ModelScope's inference API to generate images and supports LoRA. However the skill metadata declares no required environment variables or primary credential while both SKILL.md and the scripts clearly require a MODELSCOPE_API_KEY (CLI arg, env var, or ~/.modelscope/api_key). That omission in metadata is a meaningful mismatch.

Instruction Scope

SKILL.md and the included scripts provide focused runtime instructions: send generation requests to ModelScope endpoints, poll task status, download the returned image URL, and save it. The instructions tell the user to set an API key via env/CLI/config file and to pip-install requests and pillow — all within the expected scope. The scripts do write a config file under the user's home (~/.modelscope/api_key) and download remote images, which are expected for this purpose but should be noted.

Install Mechanism

There is no install spec (instruction-only skill) and included Python scripts only depend on common pip packages (requests, pillow). No binary downloads, third‑party installers, or obscure URLs are used.

Credentials

The skill requires a single API credential (MODELSCOPE_API_KEY) in practice, which is proportionate for a cloud image-generation integration — but that credential is not declared in the registry metadata (required env vars / primaryEnv are empty). The scripts also write the API key to ~/.modelscope/api_key if the user requests saving, so users should ensure the key has appropriate scope and that the file permissions are acceptable.

Persistence & Privilege

The skill does not request elevated privileges or 'always' execution. It may write a small config file (~/.modelscope/api_key) and will perform outbound network calls to api-inference.modelscope.cn and to returned image URLs. This is consistent with its purpose but means it will persist a credential if the user chooses the save-key option.

What to consider before installing

This skill's code matches its description: it calls ModelScope's inference API, polls for task completion, downloads the produced image, and can save an API key to ~/.modelscope/api_key. Before installing: (1) be aware you must provide a MODELSCOPE_API_KEY even though the registry metadata doesn't list it—check that the key has limited scope and rotate it if needed; (2) note the scripts will write that key to your home directory if you use --save-key—verify file permissions and consider providing the key via an environment variable instead; (3) the tool makes outbound HTTPS calls to api-inference.modelscope.cn and downloads remote image URLs—run in a sandbox or isolated environment if you have strong security constraints; (4) only install the declared pip dependencies from PyPI (requests, pillow) and inspect the included scripts yourself—if you want stronger assurance, run the scripts locally with a throwaway API key first. The primary issue is the metadata omission of the required env var; functionally the skill looks coherent but exercise the usual caution with API keys and outbound network behavior.

ModelScope 图片生成

调用魔搭(ModelScope) API 生成 AI 图片。支持切换模型，优化AI识别模型名称不准确。

快速开始

# 设置 API Key（首次使用）
set MODELSCOPE_API_KEY=你的密钥

# 或保存到配置文件
py scripts/generate_qwen.py --save-key 你的密钥
py scripts/generate_tongyi.py --save-key 你的密钥

# 使用z-image生成图片
py scripts/generate_tongyi.py -p "A golden cat"
py scripts/generate_tongyi.py -p "一只在月光下奔跑的银狼" -o wolf.jpg

使用qwen-image生成图片

py scripts/generate_qwen.py -p "A golden cat" py scripts/generate_qwen.py -p "一只在月光下奔跑的银狼" -o wolf.jpg

## 参数说明

| 参数 | 说明 | 默认值 |
|------|------|--------|
| `-p, --prompt` | 图片描述（必需） | - |
| `-m, --model` | 模型 ID | Tongyi-MAI/Z-Image |
| `-o, --output` | 输出文件路径 | result_image.jpg |
| `-l, --lora` | 单个 LoRA repo-id | - |
| `--lora-weight` | 单个 LoRA 权重 | 1.0 |
| `--lora-json` | 多 LoRA JSON 配置 | - |
| `--api-key` | API Key | 从环境变量/配置文件读取 |

## LoRA 支持

**使用qwen-image生成图片时，单个 LoRA:**
```bash
py scripts/generate_qwen.py -p "动漫少女" -l your-lora-id --lora-weight 0.8

使用z-image生成图片时，单个 LoRA:

py scripts/generate_tongyi.py -p "动漫少女" -l your-lora-id --lora-weight 0.8

多个 LoRA（权重之和需为 1.0）:

py scripts/generate.py -p "风景画" --lora-json "{\"lora-1\": 0.6, \"lora-2\": 0.4}"

API Key 配置

三种方式（优先级从高到低）：

1. 命令行参数: --api-key YOUR_KEY
2. 环境变量: set MODELSCOPE_API_KEY=YOUR_KEY
3. 配置文件: ~/.modelscope/api_key

获取 API Key: https://modelscope.cn/my/myaccesstoken

依赖

pip install requests pillow

支持2种常用模型，直接使用

• Qwen/Qwen-Image-2512 - 快速生成
• Tongyi-MAI/Z-Image - 快速生成（默认）
• 更多模型见 ModelScope 平台