You do not need to replace one method with the other. Move from fast reference-based exploration to deliberate modeling only when the project requires it.
Begin by defining what the model must accomplish. If it only needs to communicate a silhouette, mood, or broad design direction, an image-led starting point may be enough. If it will be manufactured, animated, measured, repeatedly edited, or integrated into a larger scene, plan for a software refinement stage from the beginning.
Next, prepare the reference rather than expecting it to solve every ambiguity. A clear subject, visible contours, and limited visual clutter make the intended form easier to interpret. When only one view is available, mark uncertain areas for later inspection instead of assuming that hidden geometry will be correct.
Then review the first result against the real requirement. Check the overall proportions, thin or occluded parts, openings, contact points, and scale. Decide whether the asset is already suitable for its purpose, needs targeted cleanup, or should be rebuilt more deliberately in a 3D package.
For refinement, preserve the useful parts of the base while correcting the risky ones. Artists may reshape major forms, rebuild surfaces, simplify topology, add thickness, establish materials, and set consistent dimensions. The goal is not to defend the first method; it is to reach a reliable asset with the least wasted effort.
Finally, document the handoff. Record the source references, known assumptions, edits made, and remaining limitations. That information helps the next person understand which parts are intentional and which parts still need validation. In this way, image-based modeling becomes an efficient front end to a broader workflow rather than a substitute for every modeling task.