Understanding the behavior of spark plasma and flame initiation in internal combustion engines leads to improvement in fuel economy and exhaust emissions. This paper experimentally investigated spark plasma stretching and cycle-to-cycle variations under various engine speed, load, and air–fuel mixtures using natural luminosity images. Natural luminosity images of combustion in an IC engine provide information about the flame speed, rate of energy release, and combustion stability. Binarization of the intensity images has been a desirable method for detecting flame front and studying flame propagation in combustors. However, binarization can cause a loss of information in the images. To study spark plasma stretching, the location of maximum intensity was tracked and compared to the trajectory of the flame centroid in binarized images as a representative for bulk flow motion. Analysis showed comparable trends between the trajectories of the flame centroid and spark stretching. From three air–fuel mixtures, the spark plasma for the lean mixture appeared to be more sensitive to the stretching. In addition, this research investigated combustion variations using two-dimensional (2D) intensity images and compared the results to coefficient of variation (COV) of indicated mean effective pressure (IMEP) computed from in-cylinder pressure data. The results revealed a good correlation between the variations of the luminosity field during the main phase of combustion and the COV of IMEP. However, during the ignition and very early flame kernel formation, utilizing the luminosity field was more powerful than in-cylinder pressure-related parameters to capture combustion variations.