ViewScale

ViewScale analyzes short-form videos (TikTok, Instagram Reels, YouTube Shorts) and explains what limits their performance, with timestamped problems and exact fixes.

Stack

Next.js 16 (App Router, TypeScript, Cache Components) · Tailwind CSS v4 · shadcn/ui (Radix) · lucide-react · Framer Motion · Recharts · Supabase (auth, Postgres, Storage) · Anthropic API (analysis) · OpenAI Whisper (transcription) · Stripe (subscriptions, test mode) · Zod

Setup

npm install
cp .env.example .env.local
npm run dev            # http://localhost:3000

Every variable is listed in .env.example. Secrets are only read on the server; nothing except NEXT_PUBLIC_* reaches the browser.

1. Supabase

Hosted project

1. Create a project at supabase.com.
2. Apply the migrations: npx supabase link --project-ref <ref> then npx supabase db push (or paste each file in supabase/migrations/ into the SQL editor, in order).
3. Put the project URL, anon key and service role key in .env.local.
4. Auth → URL Configuration: set the Site URL to your app URL and add <your-url>/auth/callback to the redirect URLs.
5. Google sign-in: Auth → Providers → Google. Create an OAuth client in Google Cloud with the redirect URI Supabase shows you, then paste the client ID and secret.
6. Uploads up to 200 MB need Supabase Pro (the Free plan caps files at 50 MB). Raise the global limit under Storage → Settings. The videos bucket itself allows 200 MB.

Local (Docker)

npx supabase start     # applies supabase/migrations and prints local keys

Put the printed API URL, anon key and service role key in .env.local. Email confirmation is off locally, so sign-up signs you straight in.

2. Anthropic and OpenAI

• ANTHROPIC_API_KEY: the analysis model defaults to claude-sonnet-5-5 (override with ANTHROPIC_MODEL).
• OPENAI_API_KEY: used for Whisper transcription. Without it, analyses still run from frames only, and the report lists the missing transcript.

3. Stripe (test mode)

1. In the Stripe dashboard (test mode), create two products with recurring monthly prices, Creator and Pro, matching lib/pricing.ts. Put the price IDs in STRIPE_PRICE_CREATOR and STRIPE_PRICE_PRO.
2. Put the secret and publishable keys in .env.local.
3. Turn on the customer portal (Settings → Billing → Customer portal) and allow switching between the two prices and cancelling.
4. Forward webhooks locally and copy the signing secret into STRIPE_WEBHOOK_SECRET:

   stripe listen --forward-to localhost:3000/api/stripe/webhook \
     --events checkout.session.completed,invoice.paid,customer.subscription.updated,customer.subscription.deleted
   

   In production, add the same endpoint and events in the dashboard.
5. Pay with test card 4242 4242 4242 4242.

4. The analysis prompt

lib/prompts/viewscale.ts holds an interim prompt. Paste the ViewScale system prompt into VIEWSCALE_SYSTEM_PROMPT, without its "Output format" section (OUTPUT_FORMAT_INSTRUCTIONS replaces it), then set PROMPT_IS_PLACEHOLDER to false. Until then, the server logs a warning on the first analysis.

Troubleshooting

"This page couldn't load" after the layout appears. A server-side part of the page threw an error. The browser only shows a generic message (often "Minified React error #441"). The real error is in the server logs: the terminal running npm run dev / npm start, or your Vercel project's Logs tab. The error ID on the page matches the digest in the log.

The usual causes, which the app now explains on screen:

• Supabase keys missing: the site works signed-out and sign-in shows "Supabase isn't connected yet". Add the keys and redeploy (NEXT_PUBLIC_* values are baked in at build time, so a redeploy is required).
• Migrations not run: the dashboard shows "Your account isn't set up in the database". Run supabase/migrations/*.sql in order, then create your account again if you signed up before running them.

Scripts

Script	What it does
npm run dev	Dev server
npm run build / npm start	Production build / server
npm run typecheck	Generates route types, then runs tsc
npm run lint	ESLint

How it works

Upload and analysis pipeline

All video processing happens in the browser, so the server needs no ffmpeg.

1. POST /api/analyses checks credits, caps uploads in progress at 3, creates the row, and returns the Storage paths.
2. The browser uploads the video straight to Supabase Storage, with progress.
3. In parallel it extracts frames with a hidden <video> + <canvas> (lib/video/frames.ts). It samples every 0.5s for the first 3s, then every 1.5s, at most 40 frames, as 512px-wide JPEGs. The first and last frames are always included. On long videos the spacing after 3s widens so the whole video fits in 40 frames.
4. It decodes the audio to 16 kHz mono WAV (lib/video/audio.ts) and uploads it.
5. POST /api/transcribe sends the WAV to Whisper and stores timestamped segments. No audio, silent audio or a Whisper failure is recorded, and the run continues from frames only.
6. POST /api/analyze takes one credit atomically (consume_credit). It then sends the labeled frames (base64), transcript, platform, context and any real performance data to Claude and saves the result. Any failure refunds the credit (refund_credit) and marks the analysis failed. The upload page offers "Retry analysis".

Progress is shown as Uploading → Extracting frames → Transcribing → Analyzing → Done, and every failure is shown on screen.

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗