Motion blur and warping appear when a fast-moving object outruns the resolution of the latent grid. The VAE compresses video into a coarse latent, so fine spatial detail is already gone before denoising begins. A fast object crosses several latent cells between frames, and the model has no representation fine enough to place it cleanly, so it smears the object across the cells it passed through. Raising output resolution helps only if the latent grid is refined accordingly; decoding a coarse latent at higher pixel resolution just enlarges the smear.
How Text-to-Video AI Works Under the Hood