Unit testing is the cornerstone of software quality assurance, yet generating effective and maintainable tests remains a long-standing challenge. Recent progress in large language models (LLMs) has created new opportunities, as these models are capable of producing tests that are syntactically correct and often highly readable. However, experience shows that LLM-based test generation still falls short in three fundamental dimensions: achieving broad coverage of program behavior, reliably revealing real defects rather than inflating coverage numbers, and producing tests that are sufficiently clear to be maintained over time. We present WiseUT, a unified framework that integrates LLMs with program analysis to provide practical, end-to-end unit test generation. WiseUT addresses the three central challenges systematically: it leverages program-analysis-guided prompting to broaden coverage, employs type-aware constraint analysis with reflective validation to reduce false positives and uncover genuine faults, and applies purification and semantic clarification to deliver clear maintainable tests. Our evaluation across real-world open-source projects and internal industrial projects shows that WiseUT consistently improves branch coverage, doubles precision in type error detection, and produces tests preferred by developers for clarity and maintainability. A demonstration video for WiseUT is available at https://www.youtube.com/watch?v=DmszXs0eEOE.