video service

About

Video and image ingestion service running on RunPod Serverless with GPU acceleration.

• Transcodes video to H.265/HEVC 1080p using NVIDIA NVENC (with CPU fallback)
• Generates thumbnails and filmstrip frames
• Converts images to WebP format
• Calculates MD5 hashes
• Uses ffmpeg with CUDA hardware acceleration for decoding and encoding

Job Types

Type	Input	Description
video	dream_uuid, extension	Full video ingestion pipeline
image	dream_uuid, extension	Image conversion to WebP
md5	dream_uuid	MD5 hash of processed video
filmstrip	dream_uuid	Generate filmstrip frames

Requirements

• Python 3.12.x
• ffmpeg
• ImageMagick (for image WebP conversion)
• NVIDIA GPU with NVENC support (optional, falls back to CPU)

Run Locally

Install dependencies

brew install ffmpeg imagemagick

Install pyenv-virtualenv (recommended)

pyenv virtualenv 3.12.2 edream_video
pyenv activate edream_video

Install requirements

pip install -r requirements.txt

Install edream SDK

git submodule init
git submodule update --remote
pip install -e python-api

Set environment variables

cp .env.example .env
# Edit .env with your BACKEND_URL and BACKEND_API_KEY

Run handler locally

python handler.py --test_input test_input.json

Edit test_input.json with a real dream_uuid to test.

Docker

Build

docker build --platform linux/amd64 -t dream-ingestion .

Run (CPU, no GPU)

docker run --platform linux/amd64 \
  -e BACKEND_URL=<backend-url> \
  -e BACKEND_API_KEY=<backend-apikey> \
  dream-ingestion \
  python3 -u /app/handler.py --test_input /app/test_input.json

Run (GPU, on NVIDIA machine)

docker run --gpus all \
  -e BACKEND_URL=<backend-url> \
  -e BACKEND_API_KEY=<backend-apikey> \
  dream-ingestion