In the field of computer graphics, simulating natural vegetation in 3D virtual scenes faces dual challenges of technological iteration and performance optimization. With the rapid development of game engine technologies, user demands for scene realism and visual immersion have expanded from geometric accuracy to dynamic interaction levels. As core elements of natural scenes, vegetation systems’ model complexity (such as morphological diversity and lighting response) and dynamic effects (like wind interaction) place high demands on rendering pipelines and computational resources. Traditional development processes require coordination across multiple stages—planning, modeling, physical simulation, and rendering—resulting in inefficiencies and fragmented visual effects. To address these issues, this paper proposes a vegetation generation paradigm that integrates procedural generation with multi-physics field coupling. Through a three-stage optimization framework, it first utilizes the SpeedTree engine’s procedural generation algorithm, combined with Perlin noise functions and physical simulation rules, to achieve parametric construction of vegetation topological structures. Concurrently, material separation and high-resolution normal mapping techniques are employed to enhance surface details. Next, a deferred decal rendering pipeline is introduced, employing instancing techniques and LOD multi-level optimization strategies to maintain rendering efficiency for hundreds of thousands of vegetation units while preserving leaf-level lighting and shadow precision. Finally, a parameterized dynamic wind field system and particle systems are implemented to achieve multi-dimensional coupling interactions among wind speed, vegetation, and terrain. Experimental results demonstrate that the procedural vegetation generation framework reconstructs the development process, compressing traditional scene construction cycles from several weeks to a few hours. It validates real-time rendering capabilities for scenes with hundreds of millions of polygons in Unreal Engine 5, significantly enhancing rendering performance compared to traditional methods. This approach provides a new technological pathway for constructing large-scale natural environments in fields such as digital twins and the meta verse.