Solo dev builds a local, hands-on tool for automating faceless video content
A desktop app called Thothium automates the production pipeline for faceless YouTube/Shorts-style content while keeping a scene-by-scene so the creator stays in control, unlike most existing tools that just take a prompt and hand back a finished black-box result. Given a topic, the pipeline runs a multi-agent chain that searches the web, drafts a script, and fact-checks it.
It then segments the script into scenes and sources visuals by pairing (FLUX) with targeted web image search. Narration comes from a locally-run voice model (Qwen3-TTS) that clones a voice from a short reference audio clip and aligns the narration down to individual words using whisper.cpp.
The final stage masters the audio (compression, high-pass filtering) with background music automatically ducked under the narration, then assembles everything through an integrated FFmpeg pipeline. Technically, every pipeline step is content-addressed via a SQLite queue and a BLAKE3 hash cache, so editing a single sentence in the script only reprocesses what actually changed rather than rerunning the whole pipeline.
Key points
- built 'Thothium', a local desktop app for automating and editing faceless video content
- Multi-agent chain (web search → draft → fact-check) automates script writing
- Combines FLUX with targeted web image sourcing per scene
- Qwen3-TTS clones a voice from a short audio sample and aligns narration word-by-word
- SQLite + BLAKE3 content-addressed caching means only changed parts get reprocessed on edits