is it lora?
this is lora or checkpoint? (Im noob) and which version use in comfy?
its a lora, v2 trained on unpruned model
what about rank? 128/256 or 512? which use?
use the unpruned model, try rank 256
i think im doing something wrong :/ nothing works for me, can you check the image?
i think im doing something wrong :/ nothing works for me, can you check the image?
I am glad I'm not the only hopeless soul here 😉
well I wouldnt use the turbo lora, all the turbo lora does is skip over the math that makes this lora even do anything. secondly use an image to animate. I dont use a turbo, I use unpruned model. I would first start by just using the PinkCherry MiniMax h3 checkpoint beta-0.6 itself and forget the lora altogether, a lora is a compressed approximation. I can give an image to Pinkcherry beta-0.6 and say "make iti do this" and it does it on the first try, ive got hundreds of videos of civit showing what it does. No turbo lora, I tried Kijai once it sucked and I deleted it. try beta-0.6 first I think as I cannot see your image, nor prompt so its really tough to say what your issue is. if youre using the lora hoping that text to video will magically make a 33B param model "fix you know what", well its not going to. YOu have to give it a first frame, or a last frame, thats what it was trained on. to text to video for you know what, you will need a fine tune model or a very good well trained lora. a newer version of pink cherry fl2va fine tune will drop soon, I would try that one first before the lora
There will be a much more powerful version of this lora comming soon, as well as the pinkcherry full fine tune checkpoint, so stay tuned. THere is also an AfterMidnight ref2va lora which is the other reference model trained on image conditioning references available v1.2