interesting, I didn't know. I was using the default 24fps that was set in a workflow and was working fine until now.

@hboxgames132 odds are that's because of error in loras you're using, if not the actual training data in Wan itself. Because everything that would be used to train is going to be something that's either 23.98, 24, 30P, 30f/60i or 60P, generally. If that's not conformed to 16fps before training then there will be distortions in Wan's output, because Wan has always been 16fps until the 5B model and ONLY the 5B model.

Most people making WFs don't understand FPS at all or half of what they do in their WFs. They just regurgitate something they didn't understand in the first place and perpetuate the screw up or misunderstanding. Few "authors" of WFs seem to have the time or interest in actually reading or understanding the models themselves, or even scanning the basic specs for the model on every model card/page for Wan posted by its own authors.

This is especially true of Youtubers. If they're yammering on about their WF, and it's a 14B WF, and they use either 24FPS or 60FPS, look elsewhere. They're idiots.

@BRai4now 

There was legitimate confusion surrounding native framerate. When 2.2 was released, I was consistently finding from sources that all models were trained on 24fps. This has since changed. I'd agree that most don't bother to take a few minutes to understand what framerate even means, working with video for the first time. But the fact remains that framerate means nothing in latent space. Not in the way we imagine it. The model is simply trained on a number of frames. It doesn't care if, say, 24 frames correspond to one second in the real world. It can only spit out what it's learned, and it can only learn over x frames. Hence the need for a native rate as a concept. Which you know, but which I've expounded upon for the general edification of no one in particular. With the addition of the caveat about the conflicting information about the native rate. Don't forget that the youtubers you mention are also thoroughly trained. They too are fed data and spit out data, albeit with somewhat less utility. How could they do otherwise? This is how we do in this glorious age.

I think anyone using 60fps straight out of decode would quickly realize their error.

Still won't make a difference to the latent. It'll just be a really quick video. That's what people don't seem to understand, the difference between framerate and frames. And the fact that the concept of 'length' only makes sense with respect to training data.

@Ponder_Stibbons and yet the model pages every where, the source for the model itself, states it's 16fps. On the same page and same list of specs where anyone would get the info about why they should care, and how the very new duel sampler workflows would work, how do you not get the fps limitation being the same as Wan2.1? They read the part about how they need to use the 2.1 VAE and how the 2.2 VAE only works with the 5B model, how do they miss this part?

Unless they never actually understood it in the 2.1 days either...where it was just as common to find totally dumbshit "60fps" workflows? These are the same people who have always been confused. And I'll admit much of this is essentially what you're saying too. It's just not that hard to find better information, better workflows, better install procedures. You just have to want to. I don't consider myself an advanced user but the signal to noise is so freakin' low sometimes and it wouldn't be so if there weren't so many grifters and "magic bullet" hopefuls.

yup, and with a 2.2 14b workflow, if you set the frame rate in the range of 16-24 with the same seed and everything else the same except the fps, it will just pack or spread apart the frames

Trust in the Semen, ask it how and it shall be revealed. The Semen always finds a way. No but seriously if you're using a lightx2 lora try using a different one. I've run into issues with some Loras were trained to work with a specific Lightx2. In my case the MoE Lightx2v worked when the others didn't . I know this is a 3 month old post but maybe someone else is having similar issues.

@burn1one1one812 thanks for responding even if it's late lol. I got it working with a different workflow truly there was a conflict of some kind, but I have no idea what it was. But anyway, now it's working