Background and Objective: Image registration is a key operation in medical image processing, allowing a plethora of applications. Mutual information (MI) is consolidated as a robust similarity metric often used for medical image registration. Although MI provides a robust registration, it usually fails when the transform needed to register an image is too big due to MI local maxima traps. Methods: In this paper, we propose and evaluate the Generalized MI (GMI), using Tsallis entropy as an affine registration metric with additive and non-additive equations. We assessed the metrics for separable affine transforms and exhaustively evaluated the GMI output signal seeking the maximum registration range through simulated gradient descent and a Monte Carlo simulation of real registrations with translated images. Results: GMI metrics showed smoother isosurfaces that better lead the algorithm to the solution. Results show significantly prolonged registration ranges, without local maxima in the metric space, evaluated through a range of 150 mm for translations, 360° for rotations, [0.5, 2] for scaling, and [-1, 1] for skewness, with a success rate of 99.99%, 88.20%, 99.99%, and 99.99%, respectively, in simulated gradient descent. We also obtained 99.75% success in the Monte Carlo simulation of 2,000 translation registrations with 1,113 double randomized subjects, using T1 and T2 brain MRI. Conclusions: The findings argue toward the reliability of GMI for long-range registrations. MI registration with Tsallis entropy requires specific entropic indexes (q) per transformation parameter and also a proper selection of additive and non-additive equation depending on the transformation type. Also, with parallel computing and more computational power, a better analysis of the signal present on images, without simplifications like voxel sampling or histogram binning, provides a more efficient and robust MI registration, as the Monte Carlo simulation shows.