This contribution presents a loop filter (ALF) technique on top of the deep learning-based in-loop filtering method in EE1-1.6. ALF is proposed to take luma samples before CNN-based filtering as the additional input where filters with 3x3 and 5x5 diamond shapes are utilized to derive the filtered samples. Compared with VTM11.0_nnvc, the proposed method with 5x5 diamond shapes reportedly shows on average {x%, x%, x%}, { x%, x%, x%}, and {x%, x%, x%} BD-rate reductions for {Y, Cb, Cr} components, under AI, RA, and LDB configurations, respectively. With 3x3 diamond shapes, the proposed method reportedly bring on average {x%, x%, x%}, { x%, x%, x%}, and {x%, x%, x%} BD-rate reductions for {Y, Cb, Cr} components, under AI, RA, and LDB configurations, respectively.