Researchers have introduced Task-Conditional Flow Matching (TCFM), a novel framework designed for the balanced adaptation of multilingual text embedding models. Recognizing that a uniform training objective often fails to capture the unique optimization requirements of various linguistic tasks, TCFM employs a specialized approach that selectively applies Flow Matching to translation-oriented tasks. In contrast, retrieval, classification, and pair-classification tasks are optimized using objectives specifically aligned with their unique learning dynamics. To ensure stable adaptation and prevent catastrophic forgetting, the framework integrates teacher-guided representation preservation alongside a structured three-stage curriculum. The effectiveness of this method was validated using the Indic Massive Text Embedding Benchmark, where TCFM established a new state-of-the-art performance level. The results demonstrate that TCFM consistently enhances embedding quality across a broad spectrum of multilingual tasks and shows robust generalization capabilities across different embedding model families. By moving away from single-objective adaptation, the TCFM methodology allows for more nuanced and task-specific feature space alignment in multilingual contexts.