The LTXVImgToVideo node does not accept height 720. When i type in 720, the node turns it into 736. The video output is 1280x736, not a common standard i suppose.
That’s normal. LTX Video works in blocks with a specific size of 32 pixels per block. And 720 does simply not fit. It rounds up to the next fitting value. Wan and quite a few other solutions shows the same behaviour. Old stable diffusion latents used to use 16 pixels.
You can crop the video afterwards. Or scale it to 1280x720.
My Wan 2.2 workflow is able to output 1280x720. Cropping afterwards is exactly what i don’t want, because cropping changes the view. The idea was to extend a video based on the last frame. Why not using Wan? The camera motion prompt obedience of Wan is … improvable.
Wan is not different, i cannot type in 720 neither. And we talk about 16 pixels in your case, the view does not change by that. This is most probably not even noticeable.
Either way, it’s simply the limit here 
Depends on the model you use. Wan 2.2 5b is limited to multiples of 32, whereas the 14b model accepts height of 720p.
Ah i see. You never stop learning