It feels like I’ve opened Pandora’s box. I’ve been training models—from Wan 2.1 --> Wan 2.2 -->LTX-2.3—specifically on scenes featuring no male characters.
Initially, the results with LTX-2.3 were terrible—transitions and movements were poor—but a chance adjustment of the parameters yielded amazing results, far surpassing what I had achieved before.
I highly recommend giving this a try; you can achieve excellent results whether you use the LTX-2.3 base model or the 10eros version.
Description
FAQ
Comments (14)
holy crap! pretty epic
Been playing around with T2V. Any way to not have the women have massive breasts?
If your input image fully exposes your breasts, it will most likely retain their original size.
Excuse me for asking, but could I know the configuration of the training parameters?
Would you mind sharing the workflow too?
You can just use the official workflow.
Thanks for creating this LoRa!
"Initially, the results with LTX-2.3 were terrible"
Would you mind sharing what parameters you changed to achieve your results?
does the thing, but audio is a bit cooked in, was having trouble trying to get it to sound different
Did you manage to get some moans or something and if so how? I just get some very harsh skin slapping sounds whatever I prompt.
@kmiki214 alas no i gave up
As with most loras, you should use "LTX2 LoRA Loader Advanced" and zero out the audio, which forces the audio to be produced by the LTX model and not the Lora.
@SPB1234T did you remove the audio aspect of the lora training? can help speed up things if you didnt
@isit Im not training a lora, you said the audio was overcooked and Im saying to get around that you can use "LTX2 LoRA Loader Advanced" to load the lora's video and not audio.
@SPB1234T Thx. Will try that.