<p>Needle-free jet injectors offer promising capabilities for drug delivery; however, achieving precise, depth-targeted penetration remains a significant challenge due to the complex interplay between formulation rheology and tissue biomechanics. This complexity precludes the use of a single equation applicable across diverse fluid classes. To address this issue, the present study employs high-speed deep tissue imaging of ex vivo porcine skin to compare the penetration behaviors of Newtonian, non-Newtonian, and protein-based formulations. Furthermore, a conceptual, data-driven prediction framework is introduced to complement scenarios where unified analytical modeling proves inadequate. High-speed near-infrared imaging, combined with optical tissue clearing techniques, was used to capture the microsecond-scale dynamics of the penetration of glycerol, carboxymethylcellulose, and bovine serum albumin solutions into ex vivo porcine skin. The experimental dataset was augmented and analyzed using five conventional machine-learning algorithms as well as a neural network model. Predictor variables included viscosity, stagnation pressure, jet velocity, Reynolds number, and fluid type. Results indicated that increasing viscosity led to reductions in jet diameter, penetration depth, and dispersion across all fluid types, albeit with distinct linear penetration sensitivities. Within the conceptual prediction framework, the multilayer perceptron neural network model demonstrated superior accuracy (<i>R</i>² = 0.85, mean absolute error = 0.13&#xa0;mm), outperforming other conventional machine learning approaches. By integrating real-tissue microsecond near-infrared visualization with a conceptual, data-driven predictive workflow, this study elucidates the factors underlying variability in penetration scaling across different fluid classes and highlights the challenges of generalizing a single global penetration equation, particularly for non-Newtonian and protein-based formulations.</p> Graphical abstract <p></p>