Korai

Korai is an AI Content Repurpose Engine. It transforms long videos into viral short clips, mindmaps, and insights automatically. Create, repurpose, and publish smarter.

Comment

Inspiration

Creators, marketers, and podcasters spend hours manually clipping and repurposing content. We wanted a tool that could make this instant turning long-form videos into viral, multilingual, and reusable assets powered by AI.

What it does

Korai transforms long-form content into post-ready viral clips with multilingual audio and captions, auto reframing, and multi–aspect ratio exports for all platforms. It also enables chat with video content, letting users extract insights to create blog posts, viral threads, newsletters, and client follow-ups making it valuable across marketing, sales, and podcast studios.

How we built it

We built Korai using Next.js, TypeScript, and Cerebras LLM APIs, orchestrating multimodal video, audio, and text pipelines. 🔗 Detailed developer docs: https://docs.korai.run/ 📺 Youtube Video Demo with architechture overview at end: YouTube Demo

Challenges we ran into

Implementing AI auto reframing was tough we finally solved it using LR-ASD and OpenCV. Synchronizing speaker audio with multilingual captions and handling MCP authentication and security for the agentic chat were also major hurdles.

Accomplishments that we're proud of

Successfully generating multilingual audio and captions for viral shorts required heavy GPU-based processing and seamless synchronization achieving this end-to-end automation was a huge milestone.

What we learned

We gained deep insights into video processing system design, FFmpeg orchestration, and multimodal pipeline optimization building scalable, efficient, and developer-friendly AI workflows.

What’s next for Korai

We’re expanding into Agentic MCP Chat with 500+ app integrations for automated workflows. Example: a travel agency or food vlogger could summarize videos from a Google Sheet → extract locations → auto-generate itineraries → and save everything to Notion. We also plan to add clip search and explore visual–language model research.

Built With

Share this project:

Updates

Submission history