In serial dependence (SD), perceptual decisions are biased toward stimuli seen in the recent past. It is believed that SD originates from a perceptual prior that the world is stable and stimuli are correlated in time. Often, however, SD is observed in conditions that deviate from this prior —i.e., in tasks involving uncorrelated stimuli. It is therefore unclear to what extent perceptual priors are crucial for SD. If there is such a prior, one would expect larger SD for “stable”, i.e., correlated, streams of stimuli but reduced or absent SD for highly variable stimuli. Here, we tested this idea by manipulating the stability of visual features in an orientation adjustment task. Twenty human observers were presented with sequences (4-12 stimuli) of brief (200 ms), low-contrast (10%), and noisy (SNR = .5) Gabor stimuli and were asked to reproduce the orientation of the last Gabor. In separate conditions, we varied the transition probabilities between the orientations in the sequence. In the Stable condition, Gabor stimuli were slightly fluctuating around a common orientation signal. In the Volatile condition, Gabor stimuli could drastically change orientation in each trial. We measured SD as the deviation of errors toward or away from the preceding orientation. We found only a weak SD in Volatile trials but no SD in the Stable condition. Rather, Stable trials revealed the build-up of adaptation-like repulsive aftereffect. Robust SD occurred only for the orientation reported in the previous trial, independently of the condition. Our results cast doubts on the involvement of perceptual priors in SD and are more supportive of SD being the byproduct of decision- and memory-related processes. Although they are typically considered a confound in SD studies, we argue that repulsive aftereffects, rather than SD, are the stereotyped and immediate reaction of our perceptual system to stable visual input.