Skip to main content
Lab Grimoire
TW EN
☕ Coffee
Spellbook / video-podcast-maker ↗ Visit project
ai-tools

video-podcast-maker

A fully automated video podcast production pipeline packaged as a Claude Code Skill. From topic input to 4K MP4 output, integrating web research, script generation, 7 TTS engines (Edge/Azure/ElevenLabs/OpenAI/Doubao/CosyVoice/Google), Remotion rendering, and support for Bilibili/YouTube and three other platforms. Includes a visual design learning system.

Best For

Anyone who wants an AI agent to automatically generate video podcasts from topic input to finished file. Specifically:

  • Researchers or writers who want to turn science communication and research results into video, but don't have time for video editing
  • Creators who regularly produce content for Bilibili or YouTube and want to automate the repetitive production workflow
  • Developers who want structured video production via Remotion React components rather than pure AI-image-stitching
  • Content teams with multilingual TTS needs (bilingual, multiple voice options)

If you only occasionally need to record a video, manual recording is faster and more natural. If your core problem is "I have a lot of content I want to turn into video, but production is too time-consuming," this skill exists to solve that scale problem.

How I Actually Use It

Most "AI video generation" tools address the asset problem (generating images or animations). video-podcast-maker addresses the pipeline problem. From "I have a topic" to "I have an uploadable video," every step in between is automated.

The depth of this pipeline is worth examining:

  1. Research: The agent conducts web research automatically — you do not paste content manually
  2. Script: Generates bilingual scripts adjusted for the target platform's style
  3. TTS: 7 engine options from free (Edge TTS) to high-quality (ElevenLabs, Azure) — select based on budget and quality requirements
  4. Rendering: Remotion (React-based video composition), not FFmpeg stitching — a structured layout system
  5. Output: 4K MP4 optimized for each platform's specifications

learn_design.py is a secondary subsystem worth looking at: it extracts design patterns from screenshots, local videos, or URLs and stores them in a searchable reference library. Subsequent videos can then automatically reuse preferred visual styles.

The OpenClaw Skill specification is complete (allowed-tools, dependencies, metadata.openclaw), directly compatible with the TVW stack (Python3 + FFmpeg + Node + npx).

image

Where It Is Strong

  • Genuinely end-to-end: From topic to MP4, research, script, TTS, and rendering are all automated
  • 7 TTS engines from free Edge TTS to ElevenLabs high-quality voice. The same pipeline supports different budget levels.
  • Remotion structured rendering: React-based layout system with real typographic layout capability, not image stitching
  • learn_design.py extracts design patterns from existing videos, letting new videos automatically inherit visual style
  • Bilibili, YouTube, and three other platforms with format optimization

Where It Fails

  • Complex dependency chain: Python3 + FFmpeg + Node.js + npx + TTS API keys (depending on engine). Environment setup is a high barrier for non-technical users.
  • Free Edge TTS naturalness is noticeably lower than paid options. Final video quality depends on TTS budget.
  • Custom layout templates require React/TypeScript familiarity.
  • Web research depth and accuracy depend on the agent's search capability. Specialized domain content may require human verification.

Pricing, Difficulty, and Risk

The skill is free and MIT licensed (with Apache-2.0 sub-components). Advanced difficulty — full functionality requires setting up multiple API keys (depending on TTS engine choice) and ensuring local FFmpeg, Node, and Python version compatibility. Using Edge TTS as a free option lowers the entry barrier.

The main risks are environment setup complexity and TTS API costs (ElevenLabs is a paid service). Verify Remotion licensing terms for commercial use.

Verdict

The right tool form for batch content video production. It does not teach you how to edit video; it lets an agent handle the entire production pipeline.

For science communicators, researchers, or creators with regular content-to-video needs, the setup cost (environment configuration + TTS keys) is one-time. After that, the human effort per video production approaches zero. If your bottleneck is production time rather than content ideas, this is infrastructure worth investing in.

Source