Inspiration
Recording a Loom-style video is easy. Editing? That’s the hard part. We’ve watched founders, teachers, and teammates burn hours re-recording just to fix one slip-up or cut filler words. Tools like ScreenStudio make things prettier, but we wanted async video to feel like editing a Google Doc that is quick, forgiving, and powered by AI.
What it does
AutoCut lets you record once and skip the re-takes. Highlight filler words like “um” and they disappear from the video. Rewrite a line and AI regenerates your voice and swaps it in. Drop in graphics or callouts with a click using Gemini Veo 3 and BytePlus. One click sends it out to TikTok, LinkedIn, or YouTube through MCP servers. Record -> clean -> publish in minutes.
How we built it
Frontend: Next.js + Tailwind for a Loom-style recorder and editor UI
Backend: FastAPI + Whisper for transcripts with word-level timestamps
Editing: FFmpeg for cutting and stitching clips, plus TTS for rewrites
Generative inserts: Veo and BytePlus APIs for animations and callouts
Publishing: Auto-formatting into vertical clips, captions, slide summaries
Connectivity: MCP and FastMCP to push content into Notion, Slack, GitHub
We wired everything so the workflow feels instant: record -> transcript -> edit -> export.
Challenges we ran into
GenAI inserts were slow and expensive. We ended up burning way more tokens than expected. Getting word-level timestamps to line up with video cuts was tricky too. It took some trial and error to get smooth edits that didn’t feel jumpy.
Accomplishments that we’re proud of
Built a working end-to-end system in < 5 hours.
Editing by text instead of timelines feels magical
Got generative inserts working inside the editor
One click publishing to TikTok, LinkedIn, YouTube
Proved that MCP and FastMCP can make AutoCut work with real-world tools
What we learned
Scope is everything.
What’s next for AutoCut
Better lip sync and voice matching for rewrites
Smarter generative inserts like branded templates or charts
More publishing presets like reels, shorts, slides, and course modules
Built With
- byteplus
- genai
- nextjs
- python
- veo3
Log in or sign up for Devpost to join the conversation.