>wasted day setting up repo for finetune>wasted another day running my first testbut we are here, I've cooked a t2v lora for cogvideox-5b, it sucked.... but I tried it with i2v model instead and well, it does turn out to have a noticeable impact on results. heres for rat sex comparisons, left is with t2v lora, right is without, no cherrypicking these are just the results so you can see while its not perfect, im not entirely crazy in that the results are better.
https://files.catbox.moe/d4vi0r.mp4https://files.catbox.moe/v4rulh.mp4https://files.catbox.moe/kr0w4g.mp4https://files.catbox.moe/874xpa.mp4will probably try a longer training overnight tonight and compare results. this was only 1000 steps think I'll just run 3000 like default and see what happens. goodnight for now /vtai/