Hi HN,I’m a indie developer, and I built ColorSeasonAI out of a personal curiosity that quickly turned into a bit of an engineering headache.A while ago, I noticed how popular seasonal color analysis had become. Offline consultations often cost hundreds of dollars, while online quizzes rely on subjective self-judgments that rarely seem accurate. My naive initial thought was that this should be straightforward to automate with computer vision and color theory, so I started hacking together a prototype.Once I got into the weeds, I ran into some genuine dilemmas. The biggest issue has been ambient lighting—separating true skin undertones from indoor warm bulbs or overcast shadows without overcorrecting has been really tough. On top of that, seasonal color analysis was conceived by stylists rather than color scientists, so trying to map intuitive concepts like muted contrast or warm undertones into strict mathematical spaces like CIELAB feels like trying to square a circle. Edge cases like dyed hair, colored contacts, or mixed undertones across different ethnic backgrounds also keep breaking my standard thresholds.I m sharing this here because I’d really appreciate some honest advice from engineers and CV folks. If you have thoughts on better ways to normalize lighting for skin-tone extraction, ideas on decoupling intrinsic biological tones from temporary conditions, or just general feedback on whether the model’s prediction feels right or totally off for you, I’d love to hear your thoughts.