Exploring Designers’ Perceptions and Practices of Collaborating with Generative AI

Introduction

In the context of a rapidly evolving landscape of generative AI and human-technology interaction, there exists a significant void in comprehending how human creators, generative AI agents, and various social stakeholders can engage in collaborative processes. Although the field of human-computer interaction (HCI) has delved into methodologies and strategies for enhancing human-AI interaction and co-creation, the growing autonomy and integration of generative AI in creative processes necessitate diverse perspectives from different stakeholders. Yet, the understanding of how human designers perceive and navigate their collaboration with generative AI within multi-stakeholder contexts remains limited. This study addresses this gap by concentrating on the domain of avatar design, which inherently involves the collaboration of multiple stakeholders, including designers, clients, and audiences.

Employing a qualitative interview-based approach, the study engages 21 avatar designers with varying degrees of experience in incorporating generative AI tools into their professional design workflows. The study aims to unravel the designers' perceptions of utilizing generative AI as co-creative agents within multi-stakeholder collaborations and to elucidate the practices and challenges they encounter in this context. This paper uncovers a complex landscape where avatar designers confront dilemmas when collaborating with co-creative AI agents and contributes to the understanding of human-AI co-creative experiences.

Background

Avatar design, which pertains to creating virtual representations of users in online environments, has gained significant importance due to the surge in virtual interactions. This field faces challenges characterized by the need for highly customizable avatars and the active engagement of multiple stakeholders in the design process. In both Academia and Creative Industries, people have different opinions on Emerging Generative AI. While some embrace AI tools for creative purposes, others are concerned about issues like copyright, style appropriation, and the potential loss of human creativity. Artists and designers are experimenting with various methodologies to collaborate with AI in creative endeavors spanning music, sound design, media art, and visual communication. It also highlights the need to address challenges in collaborations between artists, designers, and engineers in utilizing AI as a design material. Current Human-AI research primarily focuses on improving efficiency within creative teams, without addressing the broader design framework of existing generative AI systems that should consider the perspectives and interests of various social stakeholders.

Interview Study

In order to have an in-depth comprehension of designers' attitudes and perceptions towards co-creating with generative AI agents in their workflow, as well as their current practices and potential challenges in collaboration with generative AI systems, we conducted a qualitative interview-based study. The study involved 21 avatar designers in total who have got different levels of experience and expertise in utilizing generative AI tools in their professional design workflow.

In our qualitative interview-based study involving 21 professional avatar designers, we aimed to gain an in-depth understanding of designers' attitudes, perceptions, practices, and challenges related to co-creating with generative AI agents in their design workflow. To ensure that participants had the necessary background knowledge about popular generative AI models, we recruited designers with familiarity with Midjourney and Stable Diffusion, two widely-used generative AI models. While expertise in AI or computation literacy was not mandatory, participants were required to have a minimum of one year of professional experience in 2D avatar design across various domains, such as online social platforms, virtual worlds, animation, games, and live streaming.

Participants were recruited from Xiaohongshu, a social media and e-commerce platform, and Mihuashi, a web-based professional business platform in mainland China. Semi-structured interviews were conducted, addressing designers' perceptions, practices, and challenges of co-creating with generative AI, especially concerning potential career concerns and ethical issues. All interviews were conducted remotely and recorded in Mandarin, then transcribed for analysis.

To facilitate the interviews and involve participants with varying levels of AI co-creative experience, we created a web-based assistive prototype named "0-Sketch-Paint." This prototype aimed to inspire participants and featured a two-stage process for collaboration between human designers and AI: the 0-sketch process and the sketch-paint process, which aligned with existing artistic workflows. The Anything-v3.0 model was used in the prototype, combining stable diffusion and Low-Rank Adaptation algorithms for sketch generation and ControlNet for colorization. The prototype's graphical user interface was developed for accessibility.

Data analysis involved open coding, during which thematic categories emerged, including designers' perceptions and practices of working with generative AI. Designers expressed dilemmas in deciding when to use AI based on different stakeholders' interests, and challenges in understanding and adjusting AI outputs. Finally, the researchers' positionality was acknowledged, emphasizing the importance of mitigating biases and exploring how AI can coexist with designers to enhance human-AI interaction in the design domain with multi-stakeholder participation.

Findings and Discussions

Our study investigates the dynamics of collaboration between human designers and generative AI in the context of 2D static cartoon avatar design. It uncovers several critical insights and challenges.

Firstly, designers perceive AI as having a fundamentally different workflow, leading to difficulties in direct collaboration due to AI's lack of intermediate stages and inability to work with layered components.

Secondly, HAI collaboration in avatar design is influenced by various stakeholders, including ethical concerns from audiences and constraints imposed by capital investors. These dynamics affect creative freedom and collaboration choices.

Thirdly, designers face dilemmas regarding AI's ability to replicate existing artistic styles, fearing a loss of their uniqueness and ethical challenges. They are also concerned about losing their personal style.

To enhance the co-creative experience, future AI tools should align better with human designers' workflow and consider multi-stakeholder dynamics. They should improve explainability and allow designers to maintain their personal art style safely. While our study focuses on 2D static cartoon avatar design, it provides valuable insights into the broader collaboration between human designers and generative AI, shedding light on its dynamics, challenges, and potential improvements in the creative industry.