This　study　focuses　on　the　application　of　generative　artificial　intelligence　in　the　field　of　cultural　creation　and　communication,　and　proposes　an　innovative　multimodal　fusion　generative　adversarial　network　algorithm　(MM-GAN).　The　algorithm　effectively　integrates　multimodal　cultural　data　such　as　text,　image,　and　audio　by　constructing　a　cultural　feature　fusion　module,　a　dynamic　adversarial　training　mechanism,　and　a　semantic　consistency　constraint　module.　The　experiment　uses　public　cultural　data　sets　and　self-built　characteristic　cultural　corpora,　covering　scenes　such　as　poetry　creation,　film　and　television　poster　design,　and　music　melody　generation.　In　the　poetry　generation　task,　the　BLEU-4　score　of　poetry　generated　by　MM-GAN　reached　0.45,　which　is　higher　than　the　0.32　of　traditional　GAN　and　other　baseline　models.　In　the　film　and　television　poster　design　experiment,　the　average　perceptual　similarity　between　the　generated　poster　and　the　target　style　is　0.88,　far　exceeding　the　comparison　algorithm.　In　terms　of　music　melody　generation,　the　average　harmonic　rationality　score　of　the　generated　melody　is　4.2　(out　of　5　points).　The　results　show　that　MM-GAN　is　significantly　superior　to　traditional　algorithms　in　terms　of　cultural　creation　quality,　style　fit　and　semantic　coherence,　and　can　effectively　assist　cultural　creation　and　dissemination,　providing　a　new　path　for　the　development　of　this　field.　©　2025　IEEE.