unfazedanomaly964 32gb

unfazedanomaly964 I wish I had one an a6000. Those are 48gb. Great for training. I have an RTX5090

unfazedanomaly964 I do not regret buying the 5090 at all. I feel like it should have been 48gb but I understand why Nvidia didn't do it, from a business perspective. Personally, I cannot justify the extra cost of buying a used a6000 for 16 more gb of vram. I would rather buy another 5090 and a bigger power supply to run 64gb.

playtime_ai  wait what???? 80 gb? i thought it was a better and optimized model ... mmm there should be a friendly version for average users i got a 5090

CharlieBrown0115 That is the requirement for the 14b model. There is a 5b model as well. I am unsure of the requirements for that version. There will soon be quantized versions too. I am sure that we will be able to run some version of it locally.

playtime_ai playtime_ai  cant wait to see , im sure you guys will nail it making new workflows and loras based in this model, we just gotta sit and wait

CharlieBrown0115 It won't happen overnight but it will happen.

playtime_ai In lord Kijai we trust

ItsThatTimeAgain Amen.

running the 14b fp8 model at 720p is doable just barely on a 5090 if you reduce your length to 97 (98% vram used). I'm sure lora training is out of the question

unfazedanomaly964 You could also run it at 480p to save some vram and then upscale... I would imagine.

playtime_ai playtime_ai  why u blocked me ???? come on man!!

what are you talking about @playtime_ai??? 80gb? 🤣 I have no issues making videos on my 3090, great quality, 81 frames, about 130seconds

phazei he's talking about training a lora.

I havent tested this lora yet, but chances are it will run on 2.2 fine, most of them do. I ignore what wan loras were intended for (t2v, i2v, 480p, 720p etc) and I use them wherever I want with wan and it usually works out.