It didn't work for me, even setting the weight to 0.5 didn't help; the sound was completely ruined. I tried turning off the Lora sound via the kj Lora node, but that didn't help either; I could only turn it off.

Try using the LTX2 Lora Loader Advanced node instead. it's kind of a pain, because I like the convenience of the multi-lora loaders for conserving canvas space, but using this node for each of your loras in a chain instead gives you control over how loras do (or don't) affect your audio.

This LTX lora loader allows you to reduce or eliminate the audio part of LORAs. https://github.com/Brojakhoeman/Loradaddyloaderltx

@MisticRain69 May I offer a suggestion? To you—or to other LoRA creators—I recommend against embedding audio into every single LoRA model. Instead, audio-specific LoRAs and action-specific LoRAs should be kept separate. Since LTX is already prone to glitches when stacking multiple LoRAs, having multiple active LoRAs that all contain audio results in a truly terrible user experience. Although it is currently possible to mitigate this issue by using nodes to lower the LoRA audio volume, the overall result remains quite poor. These are just my personal observations; I hope you will take them into consideration. Thank you.

@Silicon_Mirage  Muting works well via the LTX-2 Master Loader LoRa-Daddy

Yeah i'm having similar issues. At 0.4 i still hear some audio degradation, compared to without the lora

@MisticRain69 We can only hope that the developers will address and optimize this issue. Currently, the audio generation component is quite poor; even when using a node to reduce volume—which does offer some improvement—the resulting audio still contains a significant amount of background noise.

@NealPage Yes, this is indeed the node I am currently using. Although there has been some improvement, the audio still suffers from static and distortion.

@MisticRain69 I took a close look at the detailed specifications of various LoRAs today, and I noticed a pattern among those created by other authors:

If a LoRA is designed specifically for video content, its parameter value for audio influence is set to 0.

Conversely, if a LoRA is designed specifically for audio content, its parameter value for video influence is set to 0.

Given that so many users have reported significant audio distortion when using your LoRA, I checked its settings and discovered that your video parameter value is 1152, while your audio parameter value is 2112.

Although you mentioned in a previous reply that you had disabled audio training, I am still puzzled as to why the audio parameter value is higher than the video parameter value.

If you would like to use other authors' LoRAs as a reference for their video and audio parameter settings, you can find the "Loradaddyloaderltx" node—shared by another user in this comment section—to inspect them.

You can compare your LoRA against the following two examples:

The LoRA trained specifically for audio is named: hentai__voice_ltx23; its parameter value for video influence is 0.

The LoRA trained specifically for video is named: LTX2.3_Sott_Enhance; its parameter value for audio influence is 0.

@iodrg244 I'm a bit more bias towards this one that works really well:
https://github.com/phazei/ComfyUI-PowerLTXLoraLoaderExtra

Lets you easily sort and load many LoRAs at once.

@phazei  This one is similar to that but newer. Might be more streamlined. https://github.com/808charlie/rgthree-comfy It's worked well for me so far.

@iodrg244 Well, I made ComfyUI-PowerLTXLoraLoaderExtra so I can definitely say that other one isn't more streamlined, and being a fork, makes it basically break the original RG3, which is just a entirely bad idea for the ecosystem. I did use RG3's power lora loader as guidance. Unlike so many of the nodes out there these days, I'm not a vibe coder and have been doing it for 30 years and made sure it was coded well.

@phazei I appreciate your work! Thanks for clarifying. I'll try yours :)