Ultrafast photochemical reactions are initiated by vibronic transitions from the reactant ground state to the excited potential energy surface, directly populating excited-state vibrational modes. The primary photochemical reaction of vision, the isomerization of retinal in the protein rhodopsin, is known to be a vibrationally coherent reaction, but the Franck–Condon factors responsible for initiating the process have been difficult to resolve with conventional time-resolved spectroscopies. Here we employ experimental and theoretical 2D photon echo spectroscopy to directly resolve for the first time the Franck–Condon factors that initiate isomerization on the excited potential energy surface and track the reaction dynamics. The spectral dynamics reveal vibrationally coherent isomerization occurring on the fastest possible time scale, that of a single period of the local torsional reaction coordinate. We successfully model this process as coherent wavepacket motion through a conical intersection on a ∼30 fs time scale, confirming the reaction coordinate as a local torsional coordinate with a frequency of ∼570 cm–1. As a result of spectral features being spread out along two frequency coordinates, we unambiguously assign reactant and product states following passage through the conical intersection, which reveal the key vibronic transitions that initiate the vibrationally coherent photochemistry of vision.