Abstract: Due to the high water sensitivity of the dispersed soil, the slope is prone to deformation and failure. At present, there is a lack of overall understanding of the instability and failure characteristics and prediction of the dispersed soil slope. In this paper, six groups of model tests are designed based on fractal theory. The failure mode of dispersed soil slope is studied by close-range photogrammetry technology, and a prediction model of instability time of rainfall-induced dispersed soil slope based on roughness parameter (SRR) is proposed. The test results show that when the fractal dimension value is large (2.7687–2.8234), the progressive hydraulic erosion damage is dominant. When the fractal dimension value is small (2.4977–2.6879), it is dominated by sliding failure. When the fractal dimension value is moderate (2.6879–2.7687), it is dominated by shallow surface flow failure, which has the common characteristics of erosion and landslides. The instability characteristics of dispersed soil slope are divided into three stages: seepage and dispersed disintegration (initial erosion), fluid-solid coexistence (constant velocity erosion) and near-saturated slide (accelerated erosion). In this paper, the slope model is used and SRR is selected as the prediction index. The prediction area is determined by improving the tangent angle model, which can predict the time of instability and failure of the flow-slip slope. In this paper, it is found that the mean RMSE of the prediction results of the failure time of the sliding failure slope is 7.68, and the mean RE is 2.4%. The research results are of great significance for understanding the failure characteristics of dispersed soil slopes, and provide a new model for the accurate prediction of rainfall-induced sliding failure landslides.