[ BACK TO PROJECTS ]
Web Development

VidFactory — AI Content Agent

An AI content agent that takes a video channel from idea to published video on autopilot: ten specialist agents plan, write, storyboard, narrate, illustrate, render, quality-check, and prepare SEO for each video, with a human approval gate before anything ships. Channels carry their own audience, tone, voice, and format (16:9 or 9:16), every step is visible, editable, and costed, and it runs on your own models — OpenAI, Gemini, Cartesia, ElevenLabs, or any OpenAI-compatible endpoint.

VidFactory — AI Content Agent

Project Overview & Architecture

FACTORY is an AI content agent for video channels. You give it a channel — who it is for, what it sounds like, how often it publishes — and it plans, writes, narrates, illustrates, animates, quality-checks, and schedules complete videos. Every step is visible, editable, and costed, and nothing is published until you approve it. It is built for people who run one channel carefully and for teams that run many at once: children's story channels, explainers, documentary-style storytelling, and vertical short-form.

A video is produced by a chain of ten specialist agents — planner, writer, storyboard, voice, visuals, video production, QC, SEO, publisher, analyst. Every agent has a typed contract, every output is validated before the next step starts, and every run is written to the database as it happens. You can watch a render progress, read the script, and see exactly which provider produced which asset.

Key capabilities:

  • Channel planning: audience, tone, voice, cadence, and format (16:9 or 9:16) defined once; the factory keeps producing on schedule, deduplicating ideas and respecting each channel's timezone.
  • Bring your own models: hosted leaders (OpenAI, Gemini, Cartesia, ElevenLabs) or any endpoint that speaks the OpenAI or Anthropic API — Groq, Together, OpenRouter, DeepSeek, Mistral, xAI, Ollama, LM Studio, vLLM, LiteLLM. Keys are stored encrypted with AES-256-GCM, never returned by the API.
  • Six Remotion templates in the box: Editorial, Children's story, Educational, Storytelling, Short-form vertical, and Editorial short — each with its own colours, typography, caption style, and scene framing.
  • Preview before you render: the dashboard embeds Remotion's Player running the same composition code the renderer uses, with scene-by-scene navigation and a frame-accurate timeline.
  • Audio Studio: single-voice generation, A/B comparison of two configurations, and multi-speaker dialogue, with usage and cost tracking per provider.
  • Know what it costs: LLM tokens, image generations, speech characters, and render compute are logged per run, per agent, and per provider.
  • Humans stay in the loop: a budget guard, an approval gate before anything leaves the studio, and a one-click cancel that stops a render mid-encode.

Architecture: Next.js 16 + TypeScript + Tailwind v4 + Remotion Player (web); FastAPI + SQLAlchemy 2 + Alembic + JWT + AES-256-GCM vault (API); PostgreSQL 16 and Redis 7; Remotion media engine; Docker deployment.

Technologies Used
Next.jsTypeScriptTailwind CSSFastAPIPythonRemotionPostgreSQLRedis

Project Links