A method and apparatus of prediction for video coding are disclosed. According to the method, when the Intra Sub-Partition (ISP) mode is applied to a block, the block is partitioned into multiple sub-blocks horizontally or vertically. A target horizontal transform and a target vertical transform are determined from a candidate transform set for each of the multiple sub-blocks according to a target setting belonging to a setting group comprising unified setting and block setting. The target horizontal transform and the target vertical transform selected are then applied to each of the multiple sub-blocks. According to another method, a target horizontal and vertical transform are determined from a candidate transform set without a transform index for the current block according to unified setting, wherein the unified setting comprises two or more Intra modes or Intra-related modes.