We investigate the second-by-second impact of Fed Chair Jerome Powell's facial and vocal emotional expressions during FOMC press conferences on the market microstructure of SPY and TLT ETFs. Using unique high-frequency multimodal data and robust wild cluster bootstrap inference, we find complex and often counterintuitive effects. Contrary to standard hypotheses, negative emotional cues (facial or vocal) predict significantly lower subsequent volatility, while positive facial expressions predict higher SPY volatility. Positive vocal cues improve TLT liquidity (narrower spreads), but negative vocal cues also predict narrower spreads at longer horizons. Significant interactions between nonverbal cues and textual sentiment reveal further complexity, such as congruent positive signals predicting lower TLT depth. The Treasury market appears more sensitive than the equity market. Our results challenge simple valence-based theories for high-frequency nonverbal cues, highlighting the importance of modality, asset class, microstructure dimension, and interactions with text in shaping immediate market dynamics.