FrameForge is a research prototype that generates simple vertical videos from text using only local AI models. It runs entirely offline. No internet is required after installation. No data is sent anywhere.
This is not a professional video tool. It is a demonstration that privacy-respecting AI content generation is possible — even if limited.
You type a short prompt:
"A robot explaining photosynthesis in a garden"
The app will:
- Generate a short script (up to 5 sentences) using Phi-3-mini
- Synthesize speech using Coqui TTS (English, robotic voice)
- Generate one static image per sentence using Stable Diffusion Turbo
- Combine them into a 1080×1920 MP4 file with centered captions
Important: Each image is generated independently. Characters, clothing, and backgrounds will not match between scenes. This is not a coherent video — it is a slideshow with voiceover.
- On a modern laptop (16 GB RAM, GPU): ~12 minutes for a 30-second output
- On Raspberry Pi 5: ~90 minutes for 30 seconds, 720p only
- Output length capped at 45 seconds
Voice quality is synthetic. Captions are basic text. There are no transitions, animations, or effects.
- Not a replacement for real video editing software
- Not capable of generating consistent characters or realistic scenes
- Not suitable for professional or commercial channels
- Not a "viral video maker"
- You want to avoid cloud-based AI generators
- You need to keep your prompts and projects private
- You are a developer or student experimenting with local AI pipelines
- You value transparency over convenience
- Windows: frameforge-win.exe
- macOS (Apple Silicon): frameforge-mac-arm64.dmg
- Linux: frameforge-linux.deb
All binaries are built from this repository. No telemetry. No network calls.
MIT License. You own all generated content.