DVFix

DVFix converts Dolby Vision video to HDR10 while preserving the rest of the file as much as possible.

It is built for local media files and keeps the workflow simple:

• run it directly against a file or directory
• launch it with no input to get guided setup
• inspect what it would do before converting anything
• validate your FFmpeg/toolchain before you start a batch

dvfix.py is the main entry point. dvfix.ps1 is the Windows launcher. dvfix.sh is the Unix shell launcher.

Start Here

If you just want to run it:

Windows

.\dvfix.ps1
.\dvfix.ps1 input.mkv
.\dvfix.ps1 "D:\Videos"

Linux

sh ./dvfix.sh
sh ./dvfix.sh input.mkv
sh ./dvfix.sh /mnt/media/Movies

Running DVFix with no input starts a guided flow that can:

• start a conversion
• inspect a file or directory without converting
• run environment checks before processing media

Common Tasks

Convert a single file:

.\dvfix.ps1 input.mkv

sh ./dvfix.sh input.mkv

Inspect a file without converting:

.\dvfix.ps1 --info input.mkv

sh ./dvfix.sh --info input.mkv

Check whether your system is ready:

.\dvfix.ps1 --check

sh ./dvfix.sh --check

Process a directory recursively:

.\dvfix.ps1 "D:\Videos"

sh ./dvfix.sh /mnt/media/Movies

Create a quick Profile 5 sample:

.\dvfix.ps1 input.mkv --sample 30

sh ./dvfix.sh input.mkv --sample 30

Output Naming

By default, DVFix writes output next to the source file and adds .noDV before the extension:

movie.mkv -> movie.noDV.mkv

If you pass an output name for a single-file run, DVFix still writes it to the input directory and adds .noDV if needed.

What DVFix Does

• Profile 7: extracts the HEVC stream, removes EL/RPU with dovi_tool, then remuxes the HDR10 base layer
• Profile 8: strips Dolby Vision RPU metadata in-place with FFmpeg
• Profile 5: re-encodes the video stream to HDR10 because there is no HDR10 base layer to preserve

For non-sample workflows, DVFix preserves audio, subtitles, attachments, chapters, and container metadata.

Requirements

Required Everywhere

• Python 3
• ffmpeg and ffprobe on PATH, or placed next to dvfix.py

Required For Specific Inputs

• dovi_tool for Profile 7 inputs
• FFmpeg with the dovi_rpu bitstream filter for Profile 8 inputs
• FFmpeg with libplacebo for normal Profile 5 HDR conversion

Platform Notes

Windows:

• .\dvfix.ps1 is the easiest launcher
• Profile 5 with libplacebo requires a Vulkan runtime that provides vulkan-1.dll

Linux:

• sh ./dvfix.sh is the documented launcher
• Profile 5 with libplacebo requires a working Vulkan loader such as libvulkan.so.1
• If you are not using NVENC, choose an encoder your FFmpeg build supports, for example --encoder libx265 --preset medium

Checks And Inspection

Use --check before a big run if you want DVFix to verify the local toolchain first.

It checks for:

• ffmpeg
• ffprobe
• dovi_tool
• the selected encoder
• dovi_rpu support for Profile 8
• libplacebo support for Profile 5
• Vulkan loader availability

Use --info when you want DVFix to probe input, report the detected Dolby Vision profile, and show the planned action without writing output.

Profile 5 Notes

Profile 5 is the most demanding path.

• It must re-encode video.
• The default encoder is hevc_nvenc.
• On systems without NVENC, you should override --encoder and usually --preset.
• If libplacebo cannot run, DVFix can still continue with --p5-force-tag, but colors may be wrong.

Quick FFmpeg sanity check:

ffmpeg -hide_banner -f lavfi -i color=c=black:s=16x16:d=0.1 -vf libplacebo -frames:v 1 -f null -

If that fails with vkGetInstanceProcAddr, your FFmpeg build is not linked against the Vulkan loader.

Options

• --check Validate tools and common codec capabilities, then exit.
• --info Inspect input and print the planned action without converting anything.
• --encoder Video encoder for re-encode workflows. Default: hevc_nvenc.
• --preset Encoder preset for re-encode workflows. Default: p7.
• --cq Constant-quality value for re-encode workflows. Default: 19.
• --p5-force-tag Skip Dolby Vision processing for Profile 5 and only tag HDR10. Colors may be wrong.
• --sample N Encode only the first N seconds.
• --sample-rand N Build a sample clip from N random segments.
• --sample-seg-len N Segment length for --sample-rand. Default: 2.
• --sample-seed N Seed for repeatable random sampling.
• --replace Delete the original file after a successful conversion.
• --temp Use a custom temp directory.
• --keep-temp Keep temp files for debugging.
• --yes Skip prompts.
• --overwrite Replace an existing output file.
• --dry-run Print commands without running them.
• --no-color or --plain Disable ANSI terminal styling.

Notes

• This repo is code-only and does not include third-party binaries.
• Only single-video-stream inputs are supported.
• Sample mode is currently supported only for Profile 5.