RenderAutoEditor (MP4 → MP4)

Spring Boot backend that:

• Accepts MP4 uploads
• Removes filler words (via whisper.cpp word timestamps)
• Removes long pauses
• Outputs a cleaned MP4

Designed for Render (Basic plan) using:

• FFmpeg for audio extraction + final render
• whisper.cpp (whisper-cli) for speech-to-text timestamps
• PostgreSQL as a job queue (SKIP LOCKED + leases)
• S3-compatible storage (Cloudflare R2 recommended) for direct uploads/downloads via pre-signed URLs

How it runs

One codebase / one Docker image, two roles:

• API: APP_ROLE=api
• Worker: APP_ROLE=worker

API (async job flow)

Create job

POST /v1/jobs
Headers: Idempotency-Key: <uuid>

Optional JSON body:

{
  "pauseThresholdMs": 1500,
  "paddingMs": 120,
  "minKeepSegmentMs": 400,
  "fillerWords": ["um","uh","ah","hmm","err"],
  "languageHint": "en"
}

Response:

{
  "jobId": "…",
  "uploadUrl": "https://…",
  "commitUrl": "/v1/jobs/<jobId>/commit",
  "statusUrl": "/v1/jobs/<jobId>"
}

Upload MP4

Upload the MP4 directly to uploadUrl (HTTP PUT, content-type video/mp4).

Commit upload (enqueue)

POST /v1/jobs/<jobId>/commit → returns 202 Accepted.

Poll status + download

GET /v1/jobs/<jobId>

When status=SUCCEEDED, you’ll receive a downloadUrl to the cleaned MP4.

Local run (dev)

Prereqs:

• Java 17
• PostgreSQL
• An S3-compatible bucket (or use R2)

Run API locally:

cd render-auto-editor
mvn -q -DskipTests package
APP_ROLE=api java -jar target/render-auto-editor-0.1.0.jar

Run Worker locally (requires ffmpeg + whisper-cli in PATH):

APP_ROLE=worker java -jar target/render-auto-editor-0.1.0.jar

Docker

Build:

docker build -t render-auto-editor:local ./render-auto-editor

Run API:

docker run --rm -p 8080:8080 -e APP_ROLE=api render-auto-editor:local

Run Worker:

docker run --rm -e APP_ROLE=worker render-auto-editor:local

Required env vars (Render)

• APP_ROLE: api or worker
• DATABASE_URL (Render Postgres connection string — entrypoint converts it to the JDBC_DATABASE_* vars Spring needs)
  • Or set JDBC_DATABASE_URL, JDBC_DATABASE_USERNAME, JDBC_DATABASE_PASSWORD directly
• STORAGE_BUCKET, STORAGE_REGION, STORAGE_ENDPOINT, STORAGE_ACCESS_KEY, STORAGE_SECRET_KEY
  • If storage is left blank the app will still start; storage-dependent endpoints return 503 until configured.

Optional:

• WHISPER_MODEL_URL, WHISPER_MODEL_PATH

One-click deploy on Render (Blueprint)

This repo ships a render.yaml that creates the API web service and the Worker background service. The Postgres database is currently commented out in the blueprint so you can verify the deploy works end-to-end first; both services boot in the nodb Spring profile and only expose /healthz.

1. In Render → New → Blueprint, point it at this GitHub repo.
2. Render reads render.yaml and creates:
   • render-auto-editor-api (web)
   • render-auto-editor-worker (background worker)
3. After the first deploy, open each service → Environment and fill the storage secrets:
   • STORAGE_BUCKET, STORAGE_ENDPOINT, STORAGE_ACCESS_KEY, STORAGE_SECRET_KEY
   • For Cloudflare R2: STORAGE_REGION=auto, STORAGE_ENDPOINT=https://<accountid>.r2.cloudflarestorage.com
   • For AWS S3: STORAGE_REGION=us-east-1 (or your region), leave STORAGE_ENDPOINT blank
4. Trigger a redeploy on both services. The API exposes /healthz; storage-only endpoints work, but job creation will fail until the database is enabled.

Enabling the database later

Once you're ready to take the system fully live:

1. In render.yaml, uncomment the databases: block and the two DATABASE_URL env vars.
2. Remove the SPRING_PROFILES_ACTIVE=nodb env vars on both services.
3. Commit/push and trigger a redeploy. Render provisions the DB and rewires DATABASE_URL to both services automatically.