DiT360 Panorama Generator

通过影石DiT360生成360度全景图，自动转格式并创建查看器

Security Scan

The skill's code and instructions are consistent with its stated purpose (calling the DiT360 Hugging Face Space, converting the image, and serving a local viewer); nothing in the package demands unrelated credentials or system-wide access, but there are a few modest operational risks to be aware of.

Purpose & Capability

Name/description match the included scripts: generator.py uses gradio_client to call the Insta360 DiT360 Hugging Face Space, converts the returned image to JPG and creates a Pannellum viewer. Small mismatch: SKILL.md claims the viewer will be automatically opened in the browser, but the scripts only start a local HTTP server and print the URL (they do not actually invoke a browser).

Instruction Scope

Runtime instructions and scripts stay within the stated purpose (network call to a Hugging Face Space, image conversion, writing output files, starting a local HTTP server). The scripts do perform network calls (gradio_client to HF Space; CDN for Pannellum JS/CSS) and will download/run code from those endpoints. The generator copies the result returned by gradio_client without additional validation; this is expected for the task but worth noting.

Install Mechanism

There is no formal install spec, but the entry script recommends installing the 'uv' tool via a remote install command piped to sh: curl -LsSf https://astral.sh/uv/install.sh | sh. Piping a remote script to sh is higher-risk than using a package manager repository. Dependency installation otherwise uses uv or pip to install gradio_client. Image conversion uses macOS 'sips' (os.system call), which silently fails on non-macOS systems.

Credentials

The skill declares no required environment variables or credentials and the code does not reference environment secrets. Network access to Hugging Face and CDN resources is required and appropriate for the stated purpose.

Persistence & Privilege

always is false and the skill does not modify other skills or system-wide agent settings. It starts a local HTTP server in the output directory but does not request elevated privileges or persistent installation.

Assessment

This package appears to do what it says: call the DiT360 Hugging Face Space, save an image, convert it to JPG, and create a local Pannellum viewer. Before running: 1) Inspect the two scripts yourself (they are small and included). 2) Be cautious about running the suggested installer command (curl ... | sh) — prefer to install 'uv' from a trusted source or use pip directly. 3) Note the image conversion uses macOS 'sips' (will fail on Linux/Windows); review/replace that line if needed. 4) The tool contacts external services (Hugging Face Space and jsdelivr CDN); only use it if you trust those endpoints. 5) If you want extra safety, run the scripts in a disposable VM or container and avoid running the remote installer piped to sh.

通过影石开源的 DiT360 模型，输入文字描述即可生成 360 度全景图，并自动创建可交互查看的网页。

功能

• 🤖 调用 Hugging Face 上的 DiT360 Space 生成全景图
• 🔄 自动将 webp 转为 jpg 格式
• 🌐 创建 Pannellum 全景查看器网页
• 🚀 启动本地 HTTP 服务器，浏览器直接查看

前置要求

• Python 3.9+
• uv (Python 包管理器)
• 网络连接（访问 Hugging Face）

使用方法

快速生成

./scripts/generate.sh "sunset over ocean beach"

自定义参数

./scripts/generate.sh "cyberpunk city at night" 42 50
# 参数：描述, seed, 推理步数

输出文件

文件	说明
output/panorama_*.webp	原始生成文件
output/panorama_*.jpg	转换后的 jpg
output/viewer.html	Pannellum 查看器

查看全景图

生成完成后，脚本会自动：

1. 启动本地 HTTP 服务器（端口 8899）
2. 在浏览器中打开查看器
3. 鼠标拖动即可 360 度旋转查看

注意事项

• 首次生成需要等待 Hugging Face Space 启动（GPU 排队）
• 生成时间：30秒 ~ 3分钟（取决于排队情况）
• 图片尺寸：2048×1024（2:1 全景比例）

相关链接

• DiT360 GitHub: https://github.com/Insta360-Research-Team/DiT360
• Hugging Face Space: https://huggingface.co/spaces/Insta360-Research/DiT360