Ship faceless shorts from text without renting a GPU cluster

What it does

Short Video Maker is a local server that turns text prompts into short-form videos by orchestrating existing tools. It synthesizes English speech with Kokoro TTS, transcribes it into timed captions via Whisper, pulls background footage from Pexels, and composites everything in Remotion. The result is a rendered vertical video with narration, subtitles, and a music bed—no generative video model required.

The interesting bit

The project deliberately avoids the “AI generates pixels from scratch” approach. By treating video creation as an assembly problem rather than a hallucination problem, it keeps the whole pipeline runnable on modest hardware with no GPU. The dual MCP and REST interfaces also mean an n8n agent can commission a video as easily as a human can hit the Web UI.

Key highlights

• Runs locally with Docker or npm; the “tiny” profile targets ~3 GB RAM and no GPU.
• Exposes both a Model Context Protocol server and a REST API for automation.
• Uses Remotion to composite scenes, with fallback “joker” search terms (nature, globe, space, ocean) when Pexels returns nothing.
• Includes a browser-based Web UI for manual generation.
• Requires only a free Pexels API key; no paid third-party video generation APIs.

Caveats

• English voiceover only; Kokoro TTS does not support other languages yet.
• All background video is strictly sourced from Pexels, so your visuals are limited to stock footage.
• Windows is explicitly unsupported, and the README warns you may need to tune Remotion concurrency and cache sizes to avoid OOM crashes on resource-constrained machines.

Verdict

Ideal for developers building automated faceless-video pipelines or self-hosting content workflows without cloud GPU bills. If you need custom 3D animation, non-English narration, or original generated visuals, this is not your tool.