The　application　of　Reconfigurable　Intelligent　Surfaces　(RIS)　plays　a　crucial　role　in　B5G/6G　networks,　and　resource　allocation　for　RIS-assisted　communication　systems　is　an　important　research　topic.　Recently,　a　method　based　on　Deep　Reinforcement　Learning　(DRL)　was　proposed　for　power　and　phase-shift　matrix　allocation　in　RIS-assisted　communication　systems,　applicable　to　environments　with　hardware　defects　and　imperfect　CSI.　However,　the　majority　internet　communications　are　videos.　existing　methods　focus　on　resource　allocation　at　the　physical　layer,　without　fully　addressing　cross-layer　issues　related　to　video　transmission.　Therefore,　we　propose　cross-layer　resource　allocation　methods　for　RIS-assisted　video　communications.　We　propose　a　scheme　that　makes　the　following　contributions　with　respect　to　the　baseline　scheme:　1)　joint　source　encoder　rate　control,　beamforming,　and　RIS　phase-shift　matrix;　2)　incorporating　the　video　rate　distortion　function　parameters　of　the　application　layer　into　the　environment＇s　state.;　3)　changing　the　reward　from　data　rate　to　Peak　Signal-to-Noise　Ratio　(PSNR),　video　quality;　4)　replaces　the　random　sample　of　stochastic　policy　during　testing　stage　by　mean　at　the　testing　stage.　Simulation　results　demonstrate　that　our　proposed　scheme　has　faster　convergence　and　outperforms　the　baseline　scheme　by　3.6dB　in　PSNR　when　12　users.　　©　2025　IEEE.