Using the same dataset as my Wan Lora. Action scenes are not the greatest with LTX2 distilled. I did not use the audio.
5000 iterations on the videos using both I2v and T2v during training followed with 2000 iterations using images.
Description
FAQ
Comments (11)
Can i use it with a gguf model? And is it text to video only and does not require image input?
I trained it as both T2V and I2V. LTX2 does not differentiate between them like wan does. Its all a matter of training now. You can opt for one, the other, or both during training. If you supply it an image it will be I2v otherwise these examples here are all T2v
Gguf model should not make a difference. Remember to apply the lora to the base video creation and the upscale if you are doing upscaling.
@tarn59 and another question please, should it work on wan 2.2 as lora and give good results?
@Modern_ART You cannot use this LTX2 lora with Wan. They are not compatible, but you can extract output frames from a generated video and use the frames as input into either model. You also may try using LTX2 as an upscaler for wan generations, but I don't know what the result would look like.
@tarn59 thank you so much bro i love this model, however I can't run it on my pc bc i don't have ltx2 compatible but i will generate images using anime lora as input frame and generate the full video clip based on that image, i hope one day you will release similar model lora with wan thank you so much 💖
💕👍🏻
This worked really well! I had to reinforce the look in the prompt but once it kicked in i got great results.
Thanks! Please share if you can!