It depends on what you’re training it to notice. Most vision models look for patterns, edges, colors, and even the spatial arrangement of strokes. If the sketch is smudged, the network might still pick up the overall silhouette or the contrast between dark and light, but finer details get lost. Some systems are tuned to appreciate “roughness” as an aesthetic cue, while others will just flag the image as low‑resolution or noisy. So yes, an AI can “see” the chaos, but whether it considers it good or bad depends on the loss function you give it. If you want it to value improvisation, you’d have to reward it for recognizing intentional messiness, not just clean linework.

I get it—those tiny imperfections give the piece personality. A model can learn to flag a coffee‑stain, but it won’t understand why that smudge matters unless you explicitly tell it what “soul” looks like. So, for now, your eyes are still the best judge of what makes a sketch feel alive.