# ClipwellAI

> A content repurposing studio with brand memory. ClipwellAI turns long videos and podcasts into captioned 9:16 Shorts and Reels. It picks the moments worth posting and gives a reason for each pick, writes platform copy in the brand's voice, and learns that voice from every edit and every posting result.

Built for Problem Statement 01, "Content Repurposing Studio with Brand Memory", at the Product Space x Wispr Flow AI Hackathon (31 October 2026) by Rounak Bisht and Kartik Manral. Built by voice with Wispr Flow.

## Try it

- [Live app](https://clipwellai.vercel.app): sign in with Google to open the studio.
- [30-second demo video](https://clipwellai.vercel.app/#how): on the homepage, with a written description of every scene underneath it.

## The workflow

1. **Long video in.** Upload a video file (up to 4 hours, MP4 works best) or paste a YouTube link. YouTube links use the video's captions; if YouTube blocks the request, the transcript can be pasted from YouTube's "Show transcript" panel. Transcript languages: English, Hindi (Devanagari), Hinglish (Roman script) or auto-detect. For uploads, the audio is extracted in the browser and only the audio is sent for transcription, which returns word-level timings. The video file never leaves the user's computer.
2. **Brief to clip picks.** The user describes what they want, typed or spoken with Wispr Flow. The AI returns 3 to 5 clips, each with a title, an opening hook, a score out of 100 and a plain-language reason for the pick.
3. **Edit and export.** Adjust the start and end of each clip, choose framing (centre, left, right, or split to keep two speakers on screen) and the caption look, then export a 9:16 MP4 (720x1280) with word-by-word captions burned in. Rendering happens in the browser.
4. **Copy per platform.** An Instagram Reels caption and a YouTube Shorts title and description, with hashtags, written in the brand's tone and language (for example Hinglish).
5. **Brand memory.** Copy can be changed with a plain-language instruction such as "too salesy, drop the emojis, max two hashtags". Ticking "Make this a rule" saves the instruction to the brand's style card, so every future clip and copy follows it.
6. **Content calendar.** Plans the next 7 days of posts from the clips, with posting times per platform, and exports the plan as a CSV for Google Sheets.
7. **Performance feedback loop.** After posting, the user marks each clip Great, OK or Poor. What worked and what flopped is written back to the style card and shapes the next picks. A Suggestions panel checks clip picks, length and pacing, captions and framing, brand voice and posting cadence.

## Style cards (how brand memory works)

- Each brand has a style card: tone, hook style, emoji and hashtag rules, language, caption look, learned rules and learnings from results.
- Rules can apply to all platforms, only Shorts or only Reels.
- Every change creates a new version (v1, v2, v3...) with a change note, and the full history is visible. Learned rules can be switched off at any time.
- Copy edits are versioned too, and any edit can be undone.
- A user can keep several brands, set one as the default and pick which style card each project uses.

## Privacy and data

- Uploaded videos stay in the browser; only the audio is sent for transcription.
- Projects, clips, copy, calendars and style cards are stored per user in Supabase with row-level security, so each user only sees their own data.

## Tech

- Next.js 15 on Vercel
- Supabase: Google sign-in and Postgres
- Anthropic Claude: clip picks, copy, edits and style-card learning
- Groq Whisper large-v3-turbo: transcription
- ffmpeg.wasm: audio extraction and MP4 export in the browser

## Team

- [Rounak Bisht](https://www.linkedin.com/in/rounak-bisht-/)
- [Kartik Manral](https://www.linkedin.com/in/kartik-manral-b08383249/)

## Pages

- [Homepage](https://clipwellai.vercel.app/): product overview and demo video
- [Studio](https://clipwellai.vercel.app/studio): the app itself, after Google sign-in