1KeyTranscoder

English | 简体中文

Windows batch transcoder for camera archives. It preserves Sony and DJI metadata, repairs multi-channel audio timing, supports hardware scaling, and can verify or repair existing deliverables.

Requirements

• Windows 10/11 x64.
• Python 3.11+.
• numpy / scipy for channel sync and audio analysis.
• NVIDIA NVENC, Intel QSV, or software x265 / SVT-AV1.
• A self-contained package includes FFmpeg, NVEncC, QSVEncC and GPAC.

Quick start

python 1kt.py --input D:\footage --output D:\archive --preset hq
python 1kt.py --input D:\footage --output D:\archive --encoder nvenc
python 1kt.py --input D:\footage --output D:\archive --encoder qsv --preset uhq

Without --encoder, the backend is selected in order: NVENC -> QSV -> x265. Hardware backends never fall back to software encoding.

Inputs and backends

Input	Handling
Sony XAVC (rtmd)	Preserves rtmd, nrtm and vendor uuid boxes
DJI Action (djmd/dbgi/tmcd)	Preserves motion and timing metadata
DJI Pocket / drones	Metadata off by default; use --preserve-metadata
Other camera files	Standard transcode

Backend	Codec
nvenc, nvenc-av1	HEVC / AV1 via NVEncC
qsv, qsv-av1	HEVC / AV1 via QSVEncC
x265, svtav1	Software HEVC / AV1

Presets: uhq, hq, small, fast.

Scaling

• --resize WxH: exact output resolution (even dimensions); does not preserve the source aspect ratio.
• --scale FACTOR: multiplies the long edge and preserves the source aspect ratio (0.5, 0.25, 2, ...); output dimensions are rounded to even values.
• --resize-algo: bilinear, bicubic, lanczos; default lanczos.
• --resize and --scale are mutually exclusive. Without either option the output resolution is unchanged.
• NVENC: --vpp-resize with --output-res.
• QSV: same plus --vpp-resize-mode quality.
• x265 / SVT-AV1 default to hardware decode and scale in that path through scale_cuda or vpp_qsv; FFmpeg CPU scaling is the fallback when hardware decode is unavailable.
• --repair and --channel-sync-transparent reject both options.

python 1kt.py ... --encoder nvenc --scale 0.5     # 4K -> 1080p
python 1kt.py ... --encoder qsv --scale 2         # 1080p -> 4K
python 1kt.py ... --encoder x265 --resize 1920x1080

Audio

• 3–8 independent PCM channels are aligned automatically (48/96 kHz).
• Long takes use time-varying alignment when the delay is not constant.
• --audio-plan JSON selects, maps, orders, aligns and encodes audio.
• Compressed audio and mixed PCM/compressed files require an explicit plan.

Verification and repair

python 1kt.py ... --check basic
python 1kt.py ... --check advanced
python 1kt.py ... --check full
python 1kt.py --input D:\deliverables --output D:\fixed --repair

basic checks core metadata and stream facts; advanced adds full structural and Gyroflow checks; full adds detailed self-checks and a PSNR/SSIM sample.

Batches are resumable. Failed files are written to logs/failed_files.json for retry with --retry-list.

Tests

python tests\full_autotest.py --level unit
python tests\full_autotest.py --level toolchain
python tests\full_autotest.py --level full

Real-feature tests require testsets/ and F:\Pictures\2026; without them, environment-only checks report FAIL.

Documentation

• Release notes v0.9.12
• Hardware scaling validation
• Hardware scaling feasibility
• Audio plan guide
• x265 profile design
• Hardware decode

License

LGPL-3.0-or-later. See LICENSE and NOTICE.