TwinTrim

Landing Page
TwinTrim's Project Gallery
Video Editing Interface, with the AI agent on the right

Inspiration

As university students, we find ourselves editing videos every so often, whether it’s for presentations, side projects, or even interviews. What should be a one-minute task often ends up taking much longer, simply because we have to spend so much time relearning workflows and tools that we don’t use on a daily basis. Complex professional editors require hours to understand and use and are barely usable for casual or first-time editors. TwinTrim was created to make these difficult and time-consuming tasks faster and easier. We also took inspiration from a winning project at Hack the North that assisted audio engineers with mixing their tracks while helping novices get started in music production.

What it does

TwinTrim is an AI-powered video editor that allows users to edit videos using common language. Users can make edits via voice commands or typing instructions through a chat interface while also being able to manually control editing with a CapCut-like interface. All edits are previewed in real time, allowing users to see the changes they implemented instantly.

How we built it

TwinTrim is a full-stack web application.

🖌️ Frontend

Next.js 16: React framework for routing and server-side rendering
TypeScript: Type-safe frontend development
Tailwind CSS: Utility-first styling for rapid UI iteration

⚙️ Backend

Python + FastAPI: Handles API requests and video processing
OpenRouter: Translates natural language prompts into structured video editing actions
ElevenLabs: Enables voice input and voice responses, converting speech into editing action requests
VideoDB SDK: Provides video streaming, indexing, and storage
FFmpeg: Executes video edits such as trimming, text overlay, and audio overlay

Challenges we ran into

Using VideoDB for video storage and streaming proved challenging, as it was our first time working with a database of this type. Overall, building a full video-editing interface was completely new territory for our team, and determining how to properly structure and manage the project was one of the biggest challenges. For many of us, this was also our first hackathon, meaning some ideas may have been executed more effectively with more time. Additionally, integrating ElevenLabs presented difficulties, as the speech-to-text functionality did not perform exactly how we wanted.

Accomplishments that we're proud of

We're proud that we were able to build a functional AI-Powered video editor with basic video editing features, such as trimming clips, splicing clips, and overlays. As well as implementing Google Text-To-Speech to make the project more accessible.

What we learned

Learning about how APIS work, like OpenRouter, ElevenLabs, and Gemini.
Figuring out how VIdeoDB works as a database.
Understanding the reasoning for the limitations when deploying different LLMs and databases into our project.

What's next for TwinTrim

Include features found in more professional video editors: undo, redo, LUTs, color grading, etc.
Include AI-compatible tools like speed adjustments, transitions, effects, AI auto-cut, and smart suggestions.
Implement Gemini to be able to scrub through specific moments in YouTube videos .
Include local import library for MP4, MP3, MOV, TXT, PNG, and JPEG.
Multi-track editing.
Stronger and more accurate AI models.