For video frame interpolation (VFI), existing deeplearning-based approaches strongly rely on the ground-truth (GT) intermediate frames, which sometimes ignore the nonunique nature of motion judging from the given adjacent frames. As a result, these methods tend to produce averaged solutions that are not clear enough. To alleviate this issue, we propose to relax the requirement of reconstructing an intermediate frame as close to the GT as possible. Towards this end, we develop a texture consistency loss (TCL) upon the assumption that the interpolated content should maintain similar structures with their counterparts in the given frames. Predictions satisfying this constraint are encouraged, though they may differ from the predefined GT. Without the bells and whistles, our plug-and-play TCL is capable of improving the performance of existing VFI frameworks. On the other hand, previous methods usually adopt the cost volume or correlation map to achieve more accurate image/feature warping. However, the O(N 2 ) (N refers to the pixel count) computational complexity makes it infeasible for highresolution cases. In this work, we design a simple, efficient (O(N )) yet powerful cross-scale pyramid alignment (CSPA) module, where multi-scale information is highly exploited. Extensive experiments justify the efficiency and effectiveness of the proposed strategy.Compared to state-of-the-art VFI algorithms, our method boosts the PSNR performance by 0.66dB on the Vimeo-Triplets dataset and 1.31dB on the Vimeo90K-7f dataset. In addition, our method is easily extended to the video frame extrapolation task. Surprisingly, our extrapolation model has achieved a 0.91dB PSNR gain over FLAVR under the same experimental setting, while being 2× times smaller in terms of the model size. At last, we show that our high-quality interpolated frames are also beneficial to the development of the video super-resolution task.