8mb.local

Media Servers

Self‑hosted GPU video compressor that lets you drop a file, pick a target size (e.g., 8 MB), and get an AV1/HEVC/H.264 output with NVENC hardware acceleration or CPU fallback.

Track releases GitHub Website

Python Latest v137 · 1mo ago Security brief →

Features

NVIDIA NVENC hardware encoding with automatic CPU fallback
Supports AV1, HEVC (H.265) and H.264 via GPU or software encoders
Real‑time progress tracking, FFmpeg logs, queue management and job history

Recent releases

View all 4 releases →

v137 Mixed 1mo

Notable features

Backend responsiveness improvements: async-safe cleanup, hw-info caching, non-blocking Celery calls from API routes
Structured request logging with real 404s for API-like paths and worker task hooks
UI improvements: clearer encoder preset vs NVENC tuning distinction and Extra Quality (CQ) mode explanations

Full changelog

Highlights

SVT-AV1 in the unified FFmpeg image; preferred fast CPU AV1 path vs libaom.
Backend responsiveness: async-safe cleanup, hw-info caching, non-blocking Celery calls from API routes.
Worker: fix compress_video sys scoping (NameError in nested FFmpeg helpers).
UI: clearer encoder preset vs NVENC tuning; Extra Quality (CQ) explained; version badge aligned with v137.
Routing / logging: real 404s for API-like paths; structured request logging and worker task hooks.
Docker / ops: .env bind footgun fix, .env.example, BUILD_VERSION 137.

Docker

docker pull jms1717/8mblocal:v137
# or
docker pull jms1717/8mblocal:latest

See CHANGELOG.md for the full v137 section.

View release on GitHub

v136 Breaking risk 1mo

⚠ Upgrade required

CPU-only hosts must use `docker-compose.cpu.yml` instead of default docker-compose.yml

Breaking changes

Default docker-compose.yml requires `gpus: all` (NVIDIA GPU support)

Notable features

Optional `target_video_bitrate_kbps` parameter for fixed bitrate compression (API, batch, UI)
Max frame rate cap extended through 120 fps: 24, 25, 30, 50, 60, 72, 90, 100, 120
Portrait orientation detection improved: full ffprobe Display Matrix captured; CUDA decode avoided when rotation present for correct output dimensions

Full changelog

### Portrait / display orientation

ffprobe now uses -show_format + -show_streams (full stream JSON) so Display Matrix rotation is present for phone HEVC/MP4 files where a narrow -show_entries query omitted it.
Rotation parsing treats rotation on any side_data_list entry as authoritative (not only when side_data_type contains "display matrix").
When rotation is non-zero, the worker avoids CUDA/NVDEC decode for H.264/HEVC/AV1 paths so decoder autorotation runs; output dimensions match display size (e.g. 1080×1920) instead of coded 1920×1080 for portrait clips stored with a −90° matrix.

Target video bitrate

Optional target_video_bitrate_kbps: compress to a fixed video bitrate (API, batch upload, and UI) instead of deriving rate only from target file size.

Max frame rate (UI)

Max frame rate cap control moved to Advanced Options on the main encode page (no longer beside Resolution).
Cap choices extended through 120 fps (24, 25, 30, 50, 60, 72, 90, 100, 120), with the same list on Batch, Settings (preset defaults), and shared $lib/fpsCap parsing for localStorage.

Docker

Default docker-compose.yml requests gpus: all (NVIDIA). Hosts without GPU passthrough can use docker-compose.cpu.yml (docker compose -f docker-compose.cpu.yml up -d --build).
Docker Hub: jms1717/8mblocal:latest and jms1717/8mblocal:v136 published for this release (BUILD_VERSION 136). Pull: docker pull jms1717/8mblocal:v136 (or :latest).