jms1717/8mblocal-backend repository overview

Backend for https://github.com/JMS1717/8mb.local/⁠

⁠8mb.local – Self-Hosted GPU Video Compressor

8mb.local is a self-hosted, "fire-and-forget" web app to compress videos to a target size using NVIDIA NVENC with modern codecs (AV1/HEVC/H.264). It features a minimal SPA UI, async FastAPI backend, Celery worker, Redis broker, and real-time progress via SSE.

⁠Features

• Drop a file, pick a target size (8MB, 25MB, 50MB, 100MB, custom)
• Codecs: AV1 (av1_nvenc, default), HEVC (hevc_nvenc), H.264 (h264_nvenc)
• Audio: Opus (libopus, default), AAC
• Speed/Quality presets mapped to NVENC -preset (P1 fast … P7 best)
• ffprobe on upload for instant estimates and quality warnings
• Real-time progress + live FFmpeg log (SSE)
• GPU-accelerated FFmpeg 6.x compiled with NVENC/NPP/Opus
• Auto-cleanup of files after retention window
• Optional Basic Auth (env-controlled)

⁠Requirements

• Docker and Docker Compose
• NVIDIA GPU with NVIDIA drivers
• NVIDIA Container Toolkit installed and configured

⁠Quick Start

1. Copy .env.example to .env and adjust values.
2. Create folders: uploads/, outputs/ (already present).
3. Start services:
   • The worker image compiles FFmpeg (first build can take a while).

# Optional: Windows PowerShell
# Ensure Docker Desktop is running and NVIDIA toolkit is installed
# Run from repo root
docker compose up --build

Visit:

• Frontend: http://localhost:5173⁠
• Backend: http://localhost:8000/docs⁠ (requires AUTH if enabled)

⁠Environment Variables

• AUTH_ENABLED=true|false
• AUTH_USER, AUTH_PASS
• FILE_RETENTION_HOURS=1
• REDIS_URL=redis://redis-broker:6379/0
• BACKEND_URL=http://localhost:8000⁠ (frontend uses this)

⁠How it Works

• Upload triggers ffprobe to extract duration and bitrates.
• Target total bitrate (kbps) = (target_MB * 8192) / duration_seconds
• Target video bitrate (kbps) = total_kbps - audio_kbps (e.g., 128)
• Worker executes FFmpeg with NVENC and streams progress to Redis Pub/Sub; backend relays via SSE.

⁠Security

Enable Basic Auth by setting AUTH_ENABLED=true and specifying AUTH_USER/AUTH_PASS.

⁠Notes

• AV1 (av1_nvenc) requires recent RTX GPUs (30/40 series) and recent drivers.
• First worker build compiles FFmpeg; subsequent builds are faster.
• For production, consider HTTPS termination and hardened auth.