1. Introduction

Recently, the development of Artificial Intelligence (AI) and 3D modeling has contributed to a rapid change in various fields such as entertainment, virtual reality, and game design. Combining AI into 3D modeling in game design will potentially provide players with more personalized experiences. This will provide players with more immersive feelings and thus completely change the game design industry. Existing studies show that 3D-based games have multiple advantages. Compared with traditional 2D game environments, 3D-based games will create a positive impact on players’ interaction, enabling them to interact more with game objects and create first-person experiences [1]. By providing a real-world-like environment, players will show more interest in exploring the open world environment and connect their real-life experiences to the game. However, creating a real-world-like environment to show 3D game experiences needs a large amount of manual work. In addition, creating those complex models of the game also faces technical challenges.

To overcome those burdens, many studies have created a solution to integrate Machine Learning (ML) and Reinforcement Learning (RL) algorithms into Procedural Content Generation (PCG) methods to automatically generate new assets that are tailored to game applications [2]. In this area, researchers have explored the use of ML algorithms for generating game elements such as textures, models, and levels [3]. Meanwhile, using this method to identify suitable assets has also been proven. These studies demonstrate the possibility that ML algorithms can automate the generation of high-quality game content, which can dramatically save time for game designers on the game environment [4].

Researchers began to study the integration of ML algorithms and PCG technology to create non-text content, such as 3D comparison models for building game experiences [5]. The RL-based methods are particularly effective, and it is believed that models can be trained without collecting data in advance. The proposed method is to train RL brokers to invent new 3D models to avoid new frustration problems over time. Therefore, 3D game assets provide a user experience-centered model that can maintain player engagement and offer a better gaming experience.

Although machine learning algorithms have great potential, significant progress is still needed in developing and implementing these algorithms to generate more efficient 3D models and environments. This article aims to provide a detailed overview of the nature of 3D modeling research based on artificial intelligence, with each section focusing on specific aspects of content creation in game applications. Finally, this article explores the potential applications of these products in real-world sports development and provides guidance for future research in this field. 2. The application of AI algorithms in 3D game modelling 3. Discussion (current challenges, future prospects) 4. Conclusion.

2. The application of AI algorithms in 3D game modelling

2.1. Overview

Artificial intelligence algorithms have thoroughly revolutionized the market of 3D game modeling by introducing automation, efficiency, and creativity into processes, which were originally done manually and seen as a heavy load for designers. With the help of PCG, the core of this transformation, game creators are allowed to generate assets such as characters, environments, and levels through enhanced machine learning techniques without exhaustive human intervention, as those machine learning models learned from a huge number of datasets of existing 3D models to generate new assets with variations. By that, models created by PCG can maintain high levels of detail and realism and can be mass-generated in a short period of time. For example, generative adversarial networks (GANs) and variational autoencoders (VAEs) are commonly used to synthesize textures, shapes, and animations, enabling creators to produce diverse content customized to specific game narratives or player preferences. With this help, it can not only reduce the development time of the game but also enhance personalization, where game elements adapt in real-time to user interactions, fostering more immersive experiences. Overall, the use of AI in the 3D game industry can streamline workflows, cut costs, and open up possibilities for infinite content variety, fundamentally shifting how games are designed and experienced.

2.2. Character modelling

In character modelling, AI algorithms are well known for automating the design of avatars, non-player characters (NPCs), and creatures, which are crucial for creating immersive storytelling in 3D games. Traditional methods required artists to manually sculpt meshes, rig skeletons, and animate movements, a time-consuming task prone to inconsistencies [6]. While machine learning collects datasets of human and animal models based on anatomy, which makes it can generate a set of anatomically accurate models of living creatures in a short time [7]. For example, convolutional neural networks (CNNs) can extrapolate from 2D sketches or photographs to create 3D representations, complete with skin textures and clothing variations. Reinforcement learning further enhances this by simulating evolutionary processes, where agents iteratively refine character traits like posture, gait, and facial expressions based on feedback loops that evaluate realism and expressiveness [8]. Studies have shown that these systems can dynamically produce customized characters based on the player's behaviors during gameplay, thereby increasing player immersion, such as responding to player actions in real-time [9]. Additionally, genetic algorithms mimic natural selection to evolve unique character designs, ensuring diversity in multiplayer games to generate unique individuality for each NPC. This AI-driven approach not only accelerates production but also allows for procedural personalization, where characters evolve based on player data, creating a sense of ownership and emotional connection. By reducing the need for repetitive manual adjustments, developers can focus on higher-level creative decisions, ultimately leading to more sophisticated and responsive game worlds.

2.3. Scene modelling

Similar to character modelling, scene modelling benefits profoundly from AI, as it can create vast, complex environments based on real-world landscapes without lots of human effort. Machine learning can get sources from satellite imagery, historical data, or artist-created prototypes to synthesize terrains, foliage, and architectural elements to ensure its reality [10]. For instance, algorithms like those based on noise functions combined with deep learning can generate realistic mountains, rivers, and forests, adjusting parameters for climate, time of day, or seasonal changes [11]. Reinforcement learning agents can also optimize the playability of the game at the same time when creating realistic environments, just like generated navigable paths in open-world games or strategic cover in shooters. This is particularly useful in simulation-heavy genres like role-playing games (RPGs), where expansive worlds demand variety to prevent repetition [12]. Evaluations using metrics such as visual fidelity and computational efficiency demonstrate that AI-generated scenes often surpass manual designs in scalability, allowing for infinite procedural worlds that evolve with gameplay. This capability is crucial for virtual reality applications, where immersive environments must respond seamlessly to user movements, enhancing the sense of presence and exploration.

2.4. Level design

Level design is another critical aspect of 3D games. With the help of AI by automating the arrangement of obstacles, objectives, and interactive elements, balanced and engaging gameplay experiences are created. Machine learning models analyze player data from previous sessions to predict optimal level structures, using clustering algorithms to identify patterns in successful designs [8]. Reinforcement learning makes it possible to generate levels that maximize player retention by balancing difficulty curves and reward placements. For example, in puzzle or platformer games, AI can procedurally create branching paths that adapt to skill levels, ensuring accessibility for novices while challenging experts. Combinatorial creativity techniques merge existing assets into novel configurations, fostering innovation without starting from scratch. This approach addresses the challenge of content fatigue in long-running games by generating endless variations, such as randomized dungeons or battle arenas [13]. Metrics like player engagement time and completion rates are used to refine these algorithms iteratively, resulting in levels that feel hand-crafted yet are produced at scale. Furthermore, AI integration with game engines like Unreal or Unity allows for real-time modifications, where levels morph based on in-game events or multiplayer dynamics, adding layers of unpredictability and replayability.

2.5. Animation and asset modeling

Animation and asset integration represent another frontier where AI algorithms streamline 3D game modelling. Traditional animation involved keyframing sequences manually, but AI now generates fluid motions through motion capture data processed via recurrent neural networks (RNNs) or transformers. These models predict natural movements for characters and objects, such as walking cycles or environmental interactions, reducing artifacts and enhancing realism. Reinforcement learning optimizes animations for efficiency, training on physics simulations to ensure compliance with game rules like gravity or collisions [8]. In asset integration, AI tools harmonize disparate elements—such as importing generated models into scenes—by automatically adjusting scales, textures, and lighting for consistency. This is vital for cohesive game worlds, where mismatched assets can break immersion [14]. Hybrid systems combining supervised and unsupervised learning allow for style transfer, adapting assets to different artistic themes, like converting realistic models to cartoonish ones [6]. Practical applications in industry show that these methods cut animation time by up to 70%, enabling faster iterations and more complex narratives [15]. As games grow in scope, AI's role in seamless integration ensures that procedural elements blend effortlessly, supporting emergent gameplay where player actions influence the environment dynamically.

3. Discussion

3.1. Challenges

3.1.1. Heavy dependency on high-quality, diverse dataset

One primary challenge in applying AI algorithms to 3D game modelling is the heavy dependency on high-quality, diverse datasets for training machine learning models. Without comprehensive data encompassing various styles, cultures, and scenarios, generated assets often suffer from biases, leading to repetitive or culturally insensitive outputs [16]. For instance, if training data is predominantly sourced from Western game designs, AI may struggle to produce authentic representations of non-Western architectures or characters, limiting global applicability [17]. This issue is compounded by the scarcity of annotated 3D datasets, as creating them requires expert labeling, which is costly and time intensive. Reinforcement learning exacerbates this, as agents need extensive simulation environments to learn effectively, often resulting in overfitting to specific game types. Moreover, ethical concerns arise when datasets include copyrighted materials without proper attribution, raising legal risks for developers. Addressing this requires collaborative efforts to build open-source repositories, but current limitations hinder the scalability of AI in diverse game genres, potentially stifling innovation in niche markets like educational or serious games [18].

3.1.2. High computational demands and resource constraints

Computational demands pose another significant limitation, as advanced AI models for 3D generation require substantial processing power and memory, often beyond the reach of small studios or individual creators. Training deep neural networks for procedural content can take days on high-end GPUs, and real-time inference in games adds latency, degrading user experience in fast-paced titles. Reinforcement learning, with its iterative trial processes, consumes even more resources, leading to high energy costs and environmental impacts. This barrier restricts adoption, favoring large companies with access to cloud infrastructure, while indie developers resort to simplified models that compromise quality. Additionally, optimizing for mobile platforms exacerbates these issues, as devices have limited hardware, forcing trade-offs between detail and performance. Solutions like model compression or federated learning are emerging, but they introduce complexities in maintaining accuracy, underscoring the need for more efficient algorithms to democratize AI in game modelling.

3.1.3. Inconsistent quality and artistic control issues

Ensuring consistent quality and artistic control remains a persistent challenge, as AI-generated assets can produce artifacts, inconsistencies, or outputs that deviate from intended visions. For example, generative models might create visually appealing scenes but fail to adhere to physical laws, resulting in unstable structures during gameplay. Human oversight is still necessary to curate and refine AI outputs, blending automation with manual tweaks, which undermines the goal of full efficiency. Evaluation metrics are often subjective, making it hard to quantify success beyond basic fidelity measures, and player feedback loops are slow to implement. Furthermore, AI lacks true creativity, relying on patterns from data rather than original innovation, which can lead to homogenized content across games. This limitation highlights the tension between automation and artistry, where over-reliance on AI risks diminishing the unique touch of human designers, necessitating hybrid workflows to balance strengths.

3.2. Future prospects

3.2.1. Multimodal AI integration for holistic models

Multimodal AI integration holds promising potential, combining text, image, audio, and even haptic inputs to create more holistic 3D game models. Future systems could allow designers to input natural language descriptions, supplemented by sketches or sounds, for generating fully realized assets. Advances in large language models and diffusion techniques will enable this, fostering intuitive tools that bridge creative ideas to execution. This prospect could revolutionize prototyping, accelerating iteration cycles and empowering non-technical creators.

3.2.2. Real-time adaptive generation via reinforcement learning

Real-time adaptive generation via enhanced reinforcement learning will allow games to evolve content dynamically based on player behavior, creating personalized narratives and challenges. As computing power improves, agents could simulate entire worlds on-the-fly, adapting to emotions detected via biometrics. This will enhance engagement in VR/AR, opening new genres like adaptive simulations for training or therapy.

3.2.3. Scalability to metaverses and mobile AR platforms

Scalability to emerging platforms like metaverses and mobile AR will broaden applications, with AI optimizing for low-power devices through edge computing. Future prospects include ethical AI standards, ensuring inclusive designs and sustainable practices, shaping a vibrant, innovative game ecosystem.

4. Conclusion

In summary, AI algorithms integrated with procedural content generation offer a game-changing approach to 3D modeling in video games, streamlining workflows, cutting down on manual effort, and opening doors to more dynamic, player-focused experiences. From automating character designs and scene creation to optimizing levels and animations, these technologies have demonstrated real potential to make games more immersive and scalable. Yet, as we've seen, hurdles like dataset biases, high resource needs, inconsistent outputs, and ethical dilemmas still need tackling to fully realize this promise. Looking ahead, blending multimodal AI, adaptive real-time systems, and collaborative human-AI tools could not only overcome these limitations but also expand into new areas like metaverses and AR platforms. Ultimately, by addressing these challenges head-on, the field can foster more inclusive, innovative game worlds that keep evolving with player needs and technological strides.