The brain versus vein dilemma in BOLD fMRI has spurred research towards more direct correlates of neuronal activation. Diffusion-weighted fMRI (dfMRI) emerged as a potential alternative 17 years ago. However, its signal origins and utility have been greatly debated. In this work, we combine ultra-high-gradients and spiral readout to characterize dfMRI contrast in humans in parameter spaces (TE, b-value, SNR and resolution) that have never been accessible before. Varying TE over a wide range while keeping the b-value fixed allowed us to detect significant hemodynamic contributions to dfMRI contrast at a b-value of 1200 s/mm2.