In the multimedia era, comics in multimedia while maintaining their ease of reading is a challenge. Web augmented reality (AR) technology offers a potential solution, but Web AR to recognize comic images requires high computational performance. Using the Manga 109 Dataset, 210 pages of “Nekodama” were analyzed to crop the central portions of each image ranging from 10 to 95% of the original dimensions. These cropped images were converted into natural feature tracking (NFT) using the NFT Marker Creator and tested with AR.js for loading speed and image recognition ability. A proportional relationship between crop percentage and loading time is observed with larger crops performing better despite screen shake. A crop rate of 45– 55% balances loading time and recognition capability and spends 2.5 s per image for continuous visual performance. The applications for continuous visual performance need a crop rate of over 70% to save about 1 s per image. Black-and-white manga maintains original recognition despite low information entropy, demonstrating AR.js’s effectiveness in enhancing the multimedia presentation of the manga.