v135 (2026-04-14)
Architecture: NVIDIA + CPU only
- Removed Intel QSV, VAAPI, and AMD AMF from the worker, API models, UI, and Docker image. In practice those paths did not work reliably in this app (container/driver friction, inconsistent behavior) and were a poor fit for strict target-size / rate-control encoding. The product now standardizes on NVENC when available and CPU encoders otherwise.
- Worker split into modules (
tasks.py,encoder.py,hw_detect.py,progress.py, …); non-NVIDIA hardware branches deleted. - API split into FastAPI routers under
backend-api/app/routers/; shared helpers indeps.py. - Frontend updated to match (no Intel/AMD UI); preset/codec and settings fixes.
- AV1 → NVENC: hardware AV1 decode (
av1_cuvid) is only used after a successful probe; otherwise libdav1d is used so batch jobs don’t fail when CUVID isn’t usable. - Docs:
docs/GPU_SUPPORT.mdand related updates
Git: bf7bfa5
Docker
docker pull jms1717/8mblocal:v135
docker pull jms1717/8mblocal:latestPin by digest after pull: docker image inspect jms1717/8mblocal:v135 --format '{{index .RepoDigests 0}}'