This study presents the design of Noral, a non-anthropomorphic AI Virtual Agent (VA), developed to create positive emotional responses through visual strategies traditionally applied to human-like agents. A 3D animated VA was developed, with eight visual variants systematically combining shape (rounded vs. sharp), motion (fluid vs. mechanical), and gradient color (saturated vs. muted). A quantitative study was conducted using the Self-Assessment Manikin (SAM) scale to measure Pleasure and Arousal affective reactions from participants in Portugal and Norway. Results indicated that the visual features significantly influenced the emotional reactions: rounded shapes and fluid motion were associated with higher pleasure, while sharp edges and saturated colors correlated with higher arousal. No significant cultural differences were found, suggesting that the derived design principles can be broadly applicable across diverse contexts. Furthermore, user familiarity with AI did not influence pleasure or arousal ratings. This research contributes to the limited literature on non-anthropomorphic VAs by demonstrating how shape, movement, and color can impact emotional perception in human-AI interaction design.