Thanks for training in the LTX space!
First, thank you for this, it's crazy how much faster it is to render and play with loras in LTX.
My only issue is the quality of the lora. The first one here adds quite a bit of noise to the output. Turning the char lora off and the output is clear and clean, turning it on and there is a lot of noise - easy to spot around hair details. I wonder if this is due to the resolution of the training material?
Anyway, just some minor feedback. Thanks again for all you share :)
Another thing I just noticed, is that this lora only likes to do square/vertical renders. If I do normal ltx wide screen (1920x1088), I get vertical letterboxing on the left and right sides.
thanks for the feedback, i am still new to ltx training, there are kinks to be ironed out
sadly, opposite to the generations, the trainings take a lot of time so it is not as easy to iterate and see what works or it isnt
but eventually we will get there :)
Question: Do you plan to continue training for LTX? As someone who definitely prioritizes speed, I’d be really sad for me not to see any more LoRA updates for LTX-2.3. With my system, using Minimax would mean reverting to old, low resolutions and facing computation times of over ten minutes; with LTX, on the other hand, it takes me 2 minutes to generate 5 seconds of video—a speed that is indispensable for me and my workflow. Minimax is too slow; it’s great, but that slowness literally kills my workflow. With LTX, in 15 minutes I’ve already created four or five I2V videos at 1532x992 and edited them in Premiere, whereas with Minimax, in the same amount of time, I haven't even finished a single generation at a lower resolution. It’s absolutely not feasible this way, but if no one continues training for LTX, I’ll be left with a practically abandoned model and nothing new being trained.