# Prophetero Audio Fix — Better Sound. Original Picture.
You finally got the picture right. Now adjust the sound without re-encoding the final video picture.
## Not just for H3 videos
Use supported local videos from LTX, H3, other AI generators, cameras or editors. Compatibility depends on file format and streams, not the model that created the clip. A Civitai base-model label does not limit Audio Fix to H3.
## What you can do
- Increase or decrease volume.
- Apply two-pass loudness normalization with an adjustable target.
- Replace sound using an audio file or another video's audio.
- Move audio earlier or later, add audio fades, or mute the video.
- Export processed audio as WAV.
- Compare short original/processed previews before saving the full clip.
## Keep the picture you like
Full video exports copy the original encoded video stream and verify its encoded packet payload before publishing the output. Audio is processed separately. The output container and complete file are not byte-identical to the source, but the final video picture is not re-encoded. Originals stay untouched.
Short previews use compressed picture for speed; they are not the final export.
## Included and required
Includes the Prophetero workflow/node, drag-and-drop/file selection, Windows installer, English/Hungarian guides and controls/help in EN, HU, DE, FR, ES, IT, PT and PL. English is the default.
Requires a working ComfyUI backend and its Python environment. No generation model, cloud account or GPU generation is needed. FFmpeg and Python dependencies perform local media processing; missing dependencies are separate downloads, with installer approval before installation.
1. Extract the entire ZIP and run START_HERE.cmd.
2. Select the backend actually used by your application. For SwarmUI, use its ComfyUI backend.
3. Fully restart that backend after installation; refreshing the page alone does not load a newly installed custom node.
4. Open Prophetero_Audio_Fix_GLOBAL.json, choose a short clip and compare the previews. Run/Queue saves the full output.
Final outputs: output/Prophetero/AudioFix/Final under your ComfyUI installation.
## Limits — v0.1.0
Supports one primary video stream and one mono/stereo audio stream. Extra tracks, subtitles and data streams are not included. Auto chooses MP4 or MKV; a forced incompatible MP4 fails instead of re-encoding the picture. Metadata preservation depends on the container. Browser uploads create local copies and use disk space. See README_EN.txt for duration, upload and format limits.
Free personal/commercial creative use and free sharing/modification are allowed under the included licence. Reselling the Prophetero software, paid modified copies or paid bundles containing it requires separate permission. Source-media rights remain your responsibility.
## Credits and original sources
- Media processing engine: [FFmpeg and its contributors](https://ffmpeg.org/).
- Python FFmpeg access: [imageio-ffmpeg and its contributors](https://github.com/imageio/imageio-ffmpeg).
- Local web integration: [aiohttp and its contributors](https://github.com/aio-libs/aiohttp).
- Host application: [ComfyUI and its contributors](https://github.com/Comfy-Org/ComfyUI).
- Video stream inspection/verification: [PyAV and its contributors](https://github.com/PyAV-Org/PyAV).
## Scope of the Prophetero contribution
Prophetero provides the workflow integration, controls, installer and user interface/monitor supplied in this package. The underlying media engines and models are third-party projects, credited below. No claim is made that Prophetero trained those models or authored the upstream engines. No official endorsement is implied.
## Rights and dependencies
The included Prophetero licence applies only to Prophetero-authored material. Third-party code, models, libraries and FFmpeg builds retain their own terms; Prophetero's software resale restriction does not replace or restrict their original licences. See THIRD_PARTY.md and any notices retained inside bundled components. You must have the rights needed for your source media.
Description
# Prophetero Audio Fix GLOBAL
## 0.3.0 — 2026-11-04
Consolidated update compared with the first public 0.1.0 edition.
- Added a large, separate selected-interval repair workspace from the existing node.
- Added same-video and external audio/video donor replacement, local mute, attenuation and local Gain (-36 to +24 dB).
- Added 3/5/8 separately auditionable donor suggestions using approximate level/texture matching, editable endpoints, nominal-FPS frame stepping, edge fades and donor gain.
- Added up to 50 named repairs, each 0.02–30 seconds, with independent enable switches, editing, removal, clear and 20 in-session undo steps.
- All enabled edits run together against original media, with shared donor decoders and one final lossy AAC encoding. Active target overlaps and donors overlapping active targets are refused; search excludes those targets too.
- Added Before/After at the same preview position, preserving play/pause state while matching media loads. Added previews near repair endings; previews remain up to four seconds and are not a gapless mixer.
- Added JSON project save/load, unfinished editor drafts and automatic browser drafts. Different incoming source files do not silently inherit an old plan. Invalid storage is retained and reported.
- Unapplied editor changes and stale previews block full repair export. All-off plans remain previewable, but cannot be exported as repairs.
- Existing whole-track modes, required workflow widget order and eight interface languages retained. English default. No new model, GPU or Python-package dependency.
- Final picture stream-copy/hash verification, unique outputs and source protection retained. Generation metadata carried forward where supported by the container. Audio remains re-encoded.
- Generic Windows installer includes all repair modules and backs up replaced files. Full backend restart is required.
Validation: synthetic combined repairs, shared donors, disabled/all-off plans, invalid inputs/overlaps, picture hash, duration/metadata and one final audio encode; handler and simulated-DOM checks; disposable installer/update checks. Complete live-browser acceptance and a real multi-fault listening test remain pending. No automatic perfect-repair or semantic donor-matching claim is made.
Not limited to H3: supported videos from LTX, other generators, cameras and edits can be used.
## 0.1.0 — 2026-10-30
- Initial whole-track volume, loudness normalization, timing, replacement, audio fades, mute and WAV extraction.
- Original/processed previews, final picture verification, unique exports, eight-language controls and Windows installer.