A video coding method operates by receiving data for a block of pixels to be encoded or decoded as a current block of a current picture of a video and the deriving a linear model based on luma and chroma samples corresponding to the current block, wherein the luma and chroma samples comprises samples of a plurality of luma blocks or samples of a plurality of chroma blocks. The method then applies the linear model to incoming or reconstructed luma samples of the current block to generate a chroma predictor of the current block and uses the chroma predictor to reconstruct chroma samples of the current block or to encode the current block.