You can't use a 1.3B lora for 14B, you can use a T2V lora for I2V.

I wouldn't even use 1.3B, that's a very small model.

It would be easier to help if we see the workflow and error, try to change unet to diffusion.

What I love about Civitai is that there is always someone there to say "it doesn't work!" right above a gallery of two dozen videos showing it working...

learn sequential prompting. scene 1, 00:01 move the hand down towards... scene 2, 00:02 her hand rubs...

@Polynechramorph Pretty sure this doesn't actually work. Do you have a scenario in which this works? My undersatnding is that WAN is not built to recognize that and it doesn't make sense to the technology as it's not the way that AI videos are constructed. It has no idea what a timecode means. It renders the entire video at once does not know what frame would correspond to what time code.

put it in the Lora folder, ComfyUI<Models<Loras

Usually T2V can be used as I2V. I2V loras have more trouble going the other way.

i don't know a LOT, but it may be that you need to give the sampler more steps to use. like, for a 10 second video (using image to text models and a KSampler), i have to give it 12 steps. anything less, and it's trying to move things in the image too fast (blurry), so giving it more steps to walk through allows for cleaner looking motion.