Back to home
Live · Publishing daily

The Inference — an AI content studio that runs itself

The Inference is an end-to-end autonomous video production and publishing system. On a schedule, it discovers topics, researches sources, writes structured scripts, generates multilingual voiceovers, sources B-roll, renders complete videos with captions and visual templates, and publishes them to YouTube — across multiple channels and languages, all from a single machine. Unlike tools that only script, voice, or edit, it coordinates the whole workflow from topic discovery to scheduled upload.

How it works
Step 1

Discover

Finds trending, relevant topics via web search.

Step 2

Research

Gathers facts and sources so the content holds up.

Step 3

Script

AI writes scene-by-scene scripts (NVIDIA Nemotron + Google Gemini).

Step 4

Build

Generates TTS voiceovers and auto-sources B-roll from Pexels.

Step 5

Render

Renders full videos via Remotion — 4 visual templates, 9 scene types, captions.

Step 6

Publish

Auto-publishes to YouTube with thumbnails, tags and scheduled release.

Key capabilities
Multi-channel
Runs several YouTube channels at once — for example English AI news, Kannada stories and Hindi psychology — each on its own schedule.
Multi-language
English, Hindi and Kannada content, with automatic translation between them.
Daily automation
Scheduled tasks kick off the pipeline and publish hands-off, every day.
Local dashboard
A web UI to preview, control and monitor every stage of the pipeline.
Multiple formats
Long-form 16:9, vertical 9:16 Shorts, and auto-generated thumbnails.
Runs on one machine
The whole studio operates locally — from topic discovery to scheduled upload.
Under the hood
Video engine
Remotion — React-based programmatic rendering
AI / LLM
NVIDIA Nemotron + Google Gemini
Text-to-speech
Microsoft Edge TTS + Kokoro (local)
Media
Pexels API (stock footage)
Publishing
YouTube Data API v3 (OAuth2)
Interface
React 19 + Tailwind CSS 4 dashboard

Not a tool in the pipeline — the whole pipeline

Most AI content tools handle a single step: scripting, or voice, or editing. The Inference owns the entire chain — discovery, research, scripting, voice, B-roll, rendering, thumbnails, metadata and scheduled publishing — and keeps multiple channels and languages running on their own schedules without someone driving each stage by hand.

Curious how it runs?

Whether you're interested in the system, a collaboration, or content for your own channels — get in touch and we'll show you the studio at work.

Get in touch