This article introduces a training method for image processing, recommending the use of specific large models and parameter settings to achieve the best results. The recommended model is SouthBridge_SD1.5_V1.0, with the LoRA weight set preferably between 0.6 and 0.8. Additionally, other parameter settings are provided, such as setting the clip skip to 2, setting the high-definition repair redraw amplitude between 0.2 and 0.4, and recommending the sampling method DPM 2M Karras with sampling steps suggested between 20 and 30. These recommendations aim to help users conduct image processing training more effectively.