@Irfantor12 👍 Also just FYI, they don't need to be connected- This lora is UNet only, no TE weights were trained, so nothing for the lora to patch on CLIP. Connecting them doesn't hurt, it just doesn't make any difference either. So far I haven't seen any Wan loras with TE weights.

try reducing the target size and length. set target size to like 600 and length to 48 and see if that works. then gradually increase till it breaks again. that workflow was tested on a 5090 so should run on that or any of the cloud services. other gpus 🤷‍♀️

@nikkatofly2612 👍you can definitely get good results with a 4090. in my experience with the 4090, which i was using until recently, target size = 800 and length(frames) = 112 were my magic numbers and about the max for reliable outputs, above that things can get funky like you saw. I didn't specifically test this lora on the 4090 though. there are a lot of other variables that can affect it too, depends on your system and setup. but what you described sounds like what i experience when pushing length or res too high.

@nikkatofly2612 ah yes you got it. it will not complain if the lora is not found, it'll just proceed without it.

For the "slomo" thing, I don't have a definitive answer for this, a lot of folks are saying it is the lightning lora, but for that particular workflow you dont want to just disable it because there are other settings related to it in the KSampler nodes. Check out some of the other comfy workflows for "Wan 2.2 I2V with Lora," they'll have the parameters all set up already and you can just add the footjob lora back in. (that's easier than me trying to explain which all settings to change 😆)

@nikkatofly2612 for that workflow- leave high disabled. Might want to check out some other example workflows for the lightning loras, to see how they have the high vs low ksampler settings. You'll start to see how they interact and when high hands off to low (total steps and what step to hand off, overlap, etc). Personally I always end up going back to this workflow with the low lightning only, but it's worth checking out how some of the others are set up. The MoE thing was new to Wan 2.2 and has only been out a couple months now, I don't follow the community closely enough to know if we've settled on a 'best practice' yet.