Heuristics for AI-driven Graphical Asset Generation Tools in Game Design and Development Pipelines: A User-Centred Approach
Kaisei Fukaya, Damon Daylamani-Zad, Harry Agius · 2025
Read Heuristics for AI-driven Graphical Asset Generation Tools in Game Design and Development Pipelines: A User-Centred Approach by Kaisei Fukaya, Damon Daylamani-Zad, Harry Agius (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-02703.txt.
Introduction
Fukaya, Daylamani-Zad, and Agius study how game makers use AI tools that generate art assets, in a 2025 paper. Generative tools promise faster production of game graphics, yet little was known about where they fit in real pipelines. The team ran a user study with 16 game designers and developers. They watched how participants created assets and what frustrated or helped them.
Everyone preferred the tools at the early design stage, churning out many rough variants and polishing the best by hand. Participants pressed for tighter links with their usual software and for outputs in standard, editable formats. From these findings the authors draft a set of heuristics for tool builders. The message is that user needs, not model power alone, decide whether such tools earn a place.
Catalog record
| CREATOR | Kaisei Fukaya, Damon Daylamani-Zad, Harry Agius |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-02703
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Heuristics for AI-driven Graphical Asset Generation Tools in Game Design and Development Pipelines: A User-Centred Approach Kaisei Fukaya, Damon Daylamani-Zad, Harry Agius Abstract Graphical assets play an important role in the design and development of games. There is potential in the use of AI-driven generative tools, to aid in creating graphical assets, thus improving game design and development pipelines. However, there is little research to address how the generative methods can fit into the wider pipeline. There also no guidelines or heuristics for creating such tools. To address this gap we conducted a user study with 16 game designers and developers to examine their behaviour and interaction with generative tools for graphical assets. The findings highlight that early design stage is preferred by all participants. Designers and developers are inclined to use such tools for creating large amounts of variations at the cost of quality as they can improve the quality of the artefacts once they generate a suitable asset. The results also strongly raised the need for better integration of such tools in existing design and development environments and the need for the outputs to be in common data formats, to be manipulatable and smoothly integrate into existing environments. The study also highlights the requirement for further emphasis on the needs of the users to incorporate these tools effectively in existing pipelines. Informed by these results, we provide a set of heuristics for creating tools that meet the expectations and needs of game designers and developers. A Preprint Keywords Graphical game assets, Artificial Intelligence, PCG, User preference, User interface 1 Introduction While usage of procedural content generation (PCG) in games is widespread, it is disproportionately applied to certain forms of content, such as environments, levels and narratives. Games such as No Man’s Sky ( Hello Games, 2016 ) , Minecraft ( Mojang Studios, 2011 ) and Dwarf Fortress ( Bay 12 Games, 2006 ) use PCG to create new content for players to see and explore, while others use it to streamline development, such as Starfield ( Skrebels, 2023 ) in which large quantities of content are generated then refined by hand. Togelius et al. ( Togelius et al., 2011 ) introduce the distinction between online and offline forms of PCG. Online PCG occurs during the game’s runtime, while offline PCG occurs during the game’s development. In online PCG, care must be taken to ensure that the range of outcomes are within the designers intentions, this naturally limits the forms of content for which PCG can be applied in a practical sense. Most cases will therefore focus on types of content such as level-layouts, or item stats, which can easily be constrained and balanced in the interest of players. Offline PCG affords the system more freedom, as all content is curated by a human designer. This leads to further possibilities in providing designers and developers more creative control over the content produced this way. Mixed-Initiative Procedural Content Generation (MI-PCG), involves both humans and PCG systems in an iterative design process ( Liapis et al., 2016 ; Lai et al., 2020 ) . It is apparent that different generative methods and techniques are designed with different use cases in mind, each attempting to solve a problem or streamline a process within a larger creative pipeline. Practitioners use many tools during game design and production, from editor tools for game engines, to asset editing software such as Photoshop and Blender, or integrated development environments such as Visual Studio. These tools each have their place within the workflows and development pipelines of game creators. While these workflows look different from company to company or from one individual to another, there exists three ubiquitous stages of development: prototyping/design, production and testing ( Ramadan and Widyani, 2013 ) . All graphical assets are formulated and produced during the prototyping, design and production stages. Within this, there are many ways that PCG can help, from early inspiration and creating rough placeholders to creating fully fledged assets or remixing existing assets for variety. Each of these use cases, naturally, will have different requirements for the complexity of input, method of interaction and quality of output. However, no research has yet explored the behaviour and interaction patterns of the users, i.e. the purposes for which designers and developers use PCG tools and how they fit in their pipelines. While there is an extensive body of literature pertaining to MI-PCG there are limited case studies which examine their use and fit in design and development pipelines ( Lai et al., 2020 ) . While it is clear that MI-PCG has the potential to improve game development asset pipelines, the behaviour, opinions and preferences of game designers and developers remain, in general, unheard. Though recent work examines usage applications for generative AI systems such as ChatGPT ( Yan et al., 2024 ) among general user bases, the specific needs and requirements within the video game industry and research have not yet been addressed. Individual MI-PCG systems may be validated with user feedback ( Li et al., 2020 ; Shen et al., 2020 ; Shen and Chen, 2021 ) , but feedback is limited to specific purposes of use, and are not contextualised within the larger scope of the development pipeline. The goal of this study is to obtain insights into where and how graphical asset PCG is most useful to game designers and developers. The following section, explores the current state of research in this area. Section 3 will expand on the motivation for this study and identify seven pipeline applications. Section 4 will then introduce the research questions. Section 5 will describe the study procedure, followed by sections 6 and 7 which will explain the mock-ups that were tested and the experiment results respectively. Sections 8 and 9 will discuss the findings and present our heuristics for creators of graphical asset generative tools, and further research. 2 Review of Current Research The process of creating games is complex, requiring design decisions at every level from the overall concept and gameplay down to the characters and environments. Designers and developers make use of increasingly capable creative tools to streamline this task, which they often tailor to their own needs ( Newell et al., 2021 ; Seidel et al., 2019 ) . In a qualitative study examining expectations of Finnish game developers regarding their tools, Kasurinen et al. ( Kasurinen et al., 2013 ) present some key insights. Across the seven organisations examined, developers were largely satisfied with the tools they have, though, when it comes to assets, many preferred to purchase rather than create in-house. It is found that the companies ’expect their tools to allow easy prototyping and the ability to design while implementing’ ( Kasurinen et al., 2013 ) . Many of the companies relied on third-party game engines, and thus compatibility with these engines was a large part of the consideration when selecting new tools. The vast amount of content required in video games requires an equally vast amount of time and resources to create. Generative approaches can be applied to reduce this work load ( Seidel et al., 2019 ) . It is important for designers and developers to have control over the content that