GhostFrame AI Studio：用 Ollama、Playwright 和 FFmpeg 將本地專案製作成靜音示範

What it does

GhostFrame AI Studio is a local-first tool for creating repeatable browser demos of projects you own. It combines an existing Ollama model for demo planning, Playwright for browser automation, FFmpeg for silent MP4 conversion, and Python/Pillow for PNG and GIF assets. A small PHP dashboard controls the workflow, while Docker Compose keeps the worker and dashboard reproducible.

Why open innovation matters

Open-weight models and local tools give developers control over where prompts and demo data are processed. The project can use an Ollama model running on a developer's own machine, while Playwright, FFmpeg and Pillow provide inspectable building blocks. Developers can change the model, scenario selectors and media pipeline rather than depending on a hosted black box.

How to run

1. Install Docker Desktop and start Ollama.
2. Pull a model such as qwen3.5:4b in Ollama.
3. Copy .env.example to .env.
4. Run docker compose up -d --build.
5. Open http://localhost:8088.
6. Start a local demo app, configure TARGET_URL and edit recorder/scenario.json with dummy data.
7. Run the recording and review output/.

Generated artifacts

• Silent H.264 MP4
• Browser screenshots
• SVG and PNG title card
• Animated GIF preview
• JSON run manifest

Current limitations

GhostFrame is not a universal recorder. Each application may need a custom scenario with valid selectors. Authentication, CAPTCHA, payment and destructive workflows are intentionally out of scope. The first Docker build needs internet access. Validate the generated assets and do not record real customer information.