I Built an AI Content Machine (Article-to-Video Workflow)

Categories:

Update, October 2026: The video above shows the first version of ClipFrom, which I put together with Lovable, ChatGPT and n8n. I’ve since rebuilt it from the ground up in Claude Code as a full app at clipfrom.ai. Claude agents now run the pipeline that n8n used to handle, and the tech stack below reflects the current version.

What if you could drop a blog post or article into a tool and get back a fully produced video with voiceover, visuals, and captions in minutes? That’s what ClipFrom does.

I built it from scratch to turn a written article into a ready-to-publish short-form video, chaining AI models for the script, visuals, voice and editing.

The Problem It Solves

Content creators and marketers produce a ton of written content but repurposing that into video is time-consuming and expensive. ClipFrom collapses that process into a single automated pipeline so you can scale video output without scaling your team.

How the First Version Worked

The workflow is fully automated using a multi-model AI stack. Here is what happens under the hood when you submit an article:

  1. Script generation: ChatGPT reads the article and distills it into a punchy, engaging video script.
  2. Voiceover: ElevenLabs converts the script into a natural-sounding AI voice.
  3. Visuals: Kling AI and Fal.ai generate video clips and imagery matched to the content.
  4. Assembly: n8n orchestrates the entire workflow, passing outputs between models and triggering each step in sequence.
  5. Delivery: The finished video is returned ready to publish.

How it’s built

ClipFrom.ai: from an article to a ready-to-post short

Claude writes and directs, specialist models make the voice and visuals, and you approve the script and clips before anything renders.

  • Input
  • AI step
  • Human checkpoint
  • Service
  1. Paste an article
    Input

    An article URL or raw text. Talking-head uploads go through a highlights path instead.

    • React app
  2. Write and direct
    AI step

    A Claude agent reads the article and writes five captions, starting with a hook, plus a scene direction for each clip.

    • Claude Managed Agents
    • Sonnet 4.6
  3. Edit the script
    Human checkpoint

    You edit the captions and pick caption style, transitions and AI, stock or mixed footage.

    • Caption editor
  4. Voice and b-roll
    Service

    A voiceover with word-level timing for synced captions, plus a video clip for each scene.

    • ElevenLabs
    • Kling AI
    • Pexels
  5. Review clips
    Human checkpoint

    Swap or regenerate any clip before the final render.

    • Clip review
  6. Render and post
    Service

    Remotion renders a 9:16 MP4 with captions and transitions. Download it or post straight to Instagram.

    • Remotion on Railway
    • Instagram Graph API
App
React, Vite, Tailwind on Vercel
Backend
Supabase Edge Functions, Postgres, Auth, Storage
AI
Claude Managed Agents, ElevenLabs, Kling AI
Rendering
Remotion on Railway
Business
Stripe credits

The Tech Stack Today

ClipFrom is now a React app I build and maintain in Claude Code:

  • Claude (Anthropic): Managed agents that read the article, write the captions and direct each scene
  • ElevenLabs: Voiceover with word-level timing for synced captions, plus voice cloning and transcription
  • Kling AI and Pexels: AI-generated and stock b-roll
  • Remotion: Video rendering, hosted on Railway
  • Supabase: Database, auth, storage and edge functions
  • Stripe: Credit-based subscriptions
  • Instagram Graph API: One-click publishing to a connected account

The first version used Lovable for the frontend, ChatGPT for scripts, n8n to orchestrate the models and Fal.ai for inference.

What I Took From It

Modern AI tools let one product person build what would have needed a full engineering team a couple of years ago. I scoped the problem, designed the workflow, selected the right models for each job, and shipped a working product.

The product is still in active development but the core workflow is real and it works. Watch the full video above to see it in action.

Follow along on YouTube for more builds like this.

Ask me anything! ↓