naqashafzal/AI-Content-Studio

🎬 AI Content Studio – Next‑Gen (v2.0)

What it is – An open‑source desktop/web application that lets you create videos, podcasts and short‑form clips automatically. It strings together a set of AI agents (research, script‑writing, quality‑check, etc.) and a media‑processing pipeline so you can go from a trending topic to a finished, loudness‑normalized video with captions, branding and optional B‑roll – all without manually editing.

---

Core capabilities

Feature	What you get
Autonomous AI Director	A supervisory multi‑agent system (StrategyAgent, TrendDiscoveryAgent, ResearchAgent, CreativeDirectorAgent, QualityAssuranceAgent, LearningAgent) that researches a topic, writes a script, checks factual accuracy and narrative flow, and can rewrite low‑scoring scenes before any expensive media is generated.
Interactive storyboard	Scene cards show the narration, visual prompts, framing notes and audio cues. You can lock scenes you like and ask the director to regenerate a single shot with a natural‑language instruction.
Magic AI Clipper 2.0	Turns long YouTube videos, podcasts or streams into vertical shorts. It fuses Whisper transcripts, audio‑volume peaks and OpenCV colour‑histogram cuts to find “viral moments”, tracks the active speaker, injects stock B‑roll and burns karaoke‑style captions.
Style‑cloning engine	Upload a reference video (e.g., a MrBeast or Vox clip). The system analyses shot length, cut frequency and caption animation, then creates a reusable style profile that can be applied to new content.
Non‑linear timeline & ripple editor	Text‑based editing: delete words in the transcript and the underlying video/audio trims automatically. An AI assistant also accepts plain‑English commands like “speed up pacing” or “mute this section”.
Audio mastering & Urdu support	FFmpeg EBU‑R128 loudness normalization for YouTube/TikTok standards, plus a phonetic engine that correctly pronounces Urdu and Roman‑Urdu text for TTS.
Resilient SQLite queue	Jobs are stored step‑by‑step (research → storyboard → qa_review …) so a crash or restart never loses progress.
Multi‑model gateway	One unified API layer that can talk to Google Gemini, any OpenAI‑compatible service, or a local Ollama model, with automatic fail‑over.
Auto‑pilot campaigns	Background bots can continuously scan trends and schedule content generation, useful for running a 24/7 channel.
Utility toolbox	Stand‑alone caption burner, aspect‑ratio converter (9:16, 16:9, 1:1), and a TTS generator with preset voices.

---

Architecture at a glance

• Frontend – Next.js (React) with Turbopack, vanilla CSS + glassmorphism UI, real‑time WebSocket updates for agent logs and timeline syncing.
• Backend – FastAPI + Uvicorn (Python 3.10+). SQLite + SQLAlchemy stores the job queue and checkpoints.
• Media processing – FFmpeg for all video/audio work, yt‑dlp for downloading source media, OpenAI Whisper (offline) for transcription, pysubs2 for subtitle generation, OpenCV for speaker tracking and histogram‑based cut detection.
• AI orchestration – A thin “gateway” routes prompts to the selected LLM provider (Gemini, OpenAI, Ollama) and returns responses to the agents.

---

Getting started (quick‑start)

1. Clone & install

   git clone https://github.com/naqashafzal/AI-Content-Studio.git
   cd AI-Content-Studio
   # Python env
   python -m venv .venv && source .venv/bin/activate
   pip install -r requirements.txt
   # Frontend deps
   cd web && npm install && cd ..
   

2. Run – On Windows the provided start.bat does everything (checks Python/Node/FFmpeg, launches the API on localhost:8008 and the UI on localhost:3000). On other OSes you can start the two servers in separate terminals:

   # backend
   python server/main.py   # FastAPI
   # frontend (in /web)
   npm run dev
   

3. Open http://localhost:3000 and follow the UI to create a new campaign, pick a trend, and watch the autonomous agents build your video.

---

Who will benefit?

• Content creators who want to scale short‑form video production without learning video‑editing tools.
• Marketers / agencies needing rapid, on‑brand video ads that match a specific visual style.
• Educators / explainer‑video makers looking for a hands‑free pipeline from script to narrated video.
• Developers interested in multi‑agent orchestration, multimodal media analysis, or building on top of the provided API.

---

Limitations & caveats

• The system relies on external LLM APIs (Gemini, OpenAI, Ollama). You need a valid API key for the provider you choose.
• Style‑cloning works by analysing shot‑level statistics; it does not copy copyrighted visual assets, but the output may still resemble the reference style.
• The “auto‑pilot” trend scanner is a simple scraper; it may miss niche topics or generate low‑quality ideas if the source data is noisy.
• All heavy media work (FFmpeg transcoding, Whisper transcription) runs locally, so a decent CPU/GPU and enough disk space are required for high‑resolution videos.

---

Community & support

The project is MIT‑licensed and openly developed. The author provides a Patreon, Instagram and a YouTube channel for tutorials and updates. A full test suite (pytest) reports 19/19 passing, and the repo includes scripts for queue recovery and secret handling.

---

Bottom line – AI Content Studio is a fairly complete, production‑ready stack that combines multi‑agent AI planning with a robust video‑processing pipeline, letting non‑technical creators generate polished short‑form videos with minimal manual effort.