Add LTX-2 Support (#644)

* WIP, adding support for LTX2

* Training on images working

* Fix loading comfy models

* Handle converting and deconverting lora so it matches original format

* Reworked ui to habdle ltx and propert dataset default overwriting.

* Update the way lokr saves to it is more compatable with comfy

* Audio loading and synchronization/resampling is working

* Add audio to training. Does it work? Maybe, still testing.

* Fixed fps default issue for sound

* Have ui set fps for accurate audio mapping on ltx

* Added audio procession options to the ui for ltx

* Clean up requirements
This commit is contained in:
Jaret Burkett
2026-01-13 04:55:30 -07:00
committed by GitHub
parent 6870ab490f
commit 5b5aadadb8
28 changed files with 2180 additions and 71 deletions

View File

@@ -107,6 +107,35 @@ const docs: { [key: string]: ConfigDoc } = {
</>
),
},
'datasets.do_audio': {
title: 'Do Audio',
description: (
<>
For models that support audio with video, this option will load the audio from the video and resize it to match
the video sequence. Since the video is automatically resized, the audio may drop or raise in pitch to match the new
speed of the video. It is important to prep your dataset to have the proper length before training.
</>
),
},
'datasets.audio_normalize': {
title: 'Audio Normalize',
description: (
<>
When loading audio, this will normalize the audio volume to the max peaks. Useful if your dataset has varying audio
volumes. Warning, do not use if you have clips with full silence you want to keep, as it will raise the volume of those clips.
</>
),
},
'datasets.audio_preserve_pitch': {
title: 'Audio Preserve Pitch',
description: (
<>
When loading audio to match the number of frames requested, this option will preserve the pitch of the audio if
the length does not match training target. It is recommended to have a dataset that matches your target length,
as this option can add sound distortions.
</>
),
},
'datasets.flip': {
title: 'Flip X and Flip Y',
description: (