Accurately　predicting　weekend　box-office　revenue　is　crucial　in　the　film　industry,　assisting　studios,　investors,　and　marketing　teams　in　their　decision-making.　This　paper　presents　a　Cross-modal　Transformer-based　Neural　Network　(CTNN)　model　that　uses　transformers　to　learn　and　mine　deep　features　from　movie　posters　and　reviews.　By　incorporating　multimodal　data　and　utilizing　transformers,　our　method　captures　films＇　visual　appeal　and　textual　sentiment,　comprehensively　comprehending　the　factors　influencing　box-office　performance.　Our　proposed　method　demonstrates　robust　performance　through　extensive　experiments　and　evaluation　with　a　Root　Mean　Squared　Error　(RMSE)　of　14,625.2142,　indicating　that　the　model　can　predict　revenue　accurately.　In　addition,　our　model　obtains　an　Average　Percent　Hit　Rate　(APHR)　of　83.60　%,　demonstrating　its　accuracy　in　predicting　revenue　ranges　for　a　significant　proportion　of　the　movies　in　the　Internet　Movie　Database.　Our　research　has　potential　applications　in　the　film　industry,　assisting　stakeholders　with　resource　allocation,　marketing　campaign　optimization,　and　strategic　planning　to　maximize　return　on　investment.　©　2023　ACM.