To train the High-performance Operation Point (HOP) in-loop filter in VVC, the Joint Video Exploration Team (JVET) provides a three-stage training strategy that uses datasets compressed by the HOP embedded VVC Test Model (VTM) for training in the last two stages. However, the use of the HOP-embedded VTM to compress the training set twice more brings huge time-consumption. To address this issue, we propose progressive learning based on QP distance to enhance HOP in-loop filter while accelerating the HOP training time. We adopt the progressive learning strategy based on QP distance to strengthen the HOP learning ability. Based on QP distance, the proposed method does not use the training sets compressed by the HOP-embedded VTM, thus leading to remarkable reduction of training time. Moreover, the uncompressed video frames do not contain compression artifacts, thus the direct use of the uncompressed video data as label for training is not effective in capturing the relationship between the compressed input and its uncompressed label. However, based on QP distance, the proposed method uses higher-quality (lower QP setting) compressed data as label for training rather than using the uncompressed data as label, thus strengthening the HOP learning ability of removing compression artifacts. Experimental results show that the HOP model generated by the proposed method achieves an average BD-rate gain of -8.31% (Y), -16.28% (U), and -18.27% (V) over the VTM-11.0 anchor in the All Intra (AI) configuration thanks to the progressive learning based on the QP distance. Moreover, the proposed method reduces total training time to only 10 days while the three-training strategy recommended by JVET takes about 45 days.