About audio generate
I have an audio clip of about 25 seconds containing human voice. When the audio is input as an external audio file, the first 10-15 seconds are normal, but after 15 seconds, the human voice starts to show electronic noise. When there is no external audio input, the human voice also gradually shows electronic noise in long videos.
The model is trained on shorter clips i bet, and 5-10-15 seconds is what the model is made for, as per LTX own recommendations.
You can longer, but things might start to fall apart.
I haven't noticed robotic voices the few times i did try run a bit longer, but havent done many such test runs.
What you can do is making multiple 10-15 videos, and stitch them together. For example in a video editor.
(but for that, you might want to look into custom audio workflows to keep voice consistent)
And there is the possibility of extending videos, in incremental steps (using mask). And make them longer at each run.
(i am adding a workflow for that real soon, often the good path since you can do one step at a time, and tweak each extension to be right, with prompt etc)
How about I get this issue from the start. What can be done about that? I have been no this for two days and I can't figure out why my audio is so bad. What is the word going around about the audio issues? Video quality is great.
could be you are using the model combinations wrong perhaps.
maybe a screenshot of your model loaders?
I think I just figured it out. I didn't know there was a LTX 2.3 version of the dynamic lora. I had been using the LTX 2. Now I am just going to see If there is a way to improve it. Thanks for the response tho.
yes that could be it. Even user made loras (that are not trained on audio) can cause distorted audio.
if you are using such user made loras and get audio issues if everything else is correct try the advanced lora loader from KJNodes and set the audio strength to zero in that lora loader.