Fastclip

• 

  login

Inspiration

Creating short-form content from a long video is still surprisingly manual. Creators often have to watch the entire recording again, search for the strongest moments, cut them precisely, reframe them vertically, write captions, and add subtitles.

Fastclip was created to remove the repetitive part of that workflow without removing the creator's control. Instead of automatically publishing content, Fastclip analyzes a video, proposes the most promising moments, explains why they were selected, and lets the creator choose and refine the final clip.

What it does

Fastclip turns a long video into ready-to-export vertical clips for YouTube Shorts, TikTok, and Instagram Reels.

A creator uploads a video and Fastclip:

Transcribes the audio with timestamps. Analyzes the transcript to identify three high-potential clips. Scores and explains every suggestion. Lets the creator preview and adjust the beginning and end of each clip. Generates a 9:16 MP4 with customizable subtitles. Creates a suggested caption and hashtags for the selected clip.

The core idea is simple: AI proposes, but the creator decides.

How we built it

Fastclip is built as a full-stack web application.

The front-end uses React, TypeScript, Vite, and Tailwind CSS to provide a fast editing experience with a project dashboard, clip suggestions, timeline controls, subtitle styles, and export states.

The back-end uses Python and FastAPI. Videos are processed with FFmpeg: audio is extracted for transcription, selected clips are cut, converted to a 9:16 format, and exported as H.264 MP4 files with burned-in subtitles.

For AI processing, Faster-Whisper generates timestamped transcriptions locally, while the Mistral API analyzes the transcript and returns exactly three structured clip suggestions. Each suggestion includes timestamps, a quality score, a title, a hook, an explanation, a suggested caption, and hashtags.

Fastclip is optimized for lightweight cloud infrastructure: videos are limited in duration and size, processing is queued, only one video job runs at a time, and temporary media files are automatically deleted after processing.

Challenges we ran into

The biggest challenge was making the AI suggestions useful rather than random. A good short clip needs context, a strong opening, a clear idea, and an appropriate duration. We designed the AI output as strict structured data so the application receives precise timestamps and explanations instead of unstructured text.

Video processing was another challenge. Rendering videos can be resource-intensive, especially on a small cloud machine. We reduced the workload by extracting mono 16 kHz audio for transcription, limiting upload size and duration, exporting at 720 × 1280, and processing one job at a time.

Finally, we focused on transparency. The user can see why a clip was suggested, edit it before export, and never has to accept an AI decision blindly.## Accomplishments that we're proud of

What we learned

This project taught us how to combine AI text analysis with real media-processing workflows. We learned how to build a reliable pipeline from uploaded video to transcription, structured AI output, editable clip selection, subtitles, and final export.

We also learned that an effective AI creator tool is not one that replaces creative decisions. It is one that saves time while keeping the human in control.

What's next for Fastclip

Next, we plan to scale Fastclip on Google Cloud with more powerful compute instances to support longer and higher-resolution videos. We also want to add automatic face tracking for smarter vertical framing, reusable brand presets for subtitle styles, collaborative workflows for editors and creators, and direct publishing integrations.

Built With

• fastapi
• faster-whisper
• ffmpeg
• google-cloud
• mistral-ai
• python
• react
• sqlite
• tailwindcss
• typescript
• vite