Workflow for using the starlet model as an automatic refinement of the analytical model. From a first estimation of the lens model, model residuals are assessed through visual inspection and/or goodness-of-fit statistics. If the current source model is insufficient, the lens model is fixed and the source is reconstructed using sparse optimisation with starlets. Potential sub-components of the source are located in source plane through special filtering followed by local maxima detection. A pair of profiles shapelets+Sérsic is assigned to each sub-components, with low complexity (shapelet order n_max = 2). An iterative process is engaged for refining individual shapelet orders based on residuals in pixels corresponding to their sub-component (see Appendix B for details). Once refined orders are sufficient, a full model optimisation can be performed.