For assisting human lower-limb movement, most present exoskeletons provide predefined gaits which are not natural and adaptive to different walking conditions. This study presents a novel method for generating adaptive gait for lower-limb exoskeletons using multimodal sensor bands. The multimodal sensors can capture the users' limb movements and force myography information. Through a feed-forward neural network, an adjustable gait can be generated and given to a dynamic movement primitives (DMP) model for real-time adjustments to walking trajectories. The experimental validation and metabolic evaluations confirm the effectiveness of this strategy, demonstrating enhanced walking assistance compared to standard gait models.