[Paper Review] The Creativity of Text-to-Image Generation

Jonas Oppenlaender|arXiv (Cornell University)|May 13, 2022

Data Visualization and Analytics4 citations

TL;DR

This paper argues that the traditional product-centered view of creativity fails to capture the full scope of human creativity in text-to-image generation, proposing instead a process- and context-centered approach using Rhodes' '4 P' model (product, person, process, press). It highlights prompt engineering and online communities as central to creative practice, demonstrating that creativity emerges from human-AI interaction and community-driven learning rather than just the final image output.

ABSTRACT

Text-guided synthesis of images has made a giant leap towards becoming a mainstream phenomenon. With text-to-image generation systems, anybody can create digital images and artworks. This provokes the question of whether text-to-image generation is creative. This paper expounds on the nature of human creativity involved in text-to-image art (so-called "AI art") with a specific focus on the practice of prompt engineering. The paper argues that the current product-centered view of creativity falls short in the context of text-to-image generation. A case exemplifying this shortcoming is provided and the importance of online communities for the creative ecosystem of text-to-image art is highlighted. The paper provides a high-level summary of this online ecosystem drawing on Rhodes' conceptual four P model of creativity. Challenges for evaluating the creativity of text-to-image generation and opportunities for research on text-to-image generation in the field of Human-Computer Interaction (HCI) are discussed.

Motivation & Objective

• To challenge the product-centered definition of creativity in text-to-image generation, which reduces creativity to the originality and effectiveness of the final image.
• To demonstrate that the creative process in text-to-image art is not captured by evaluating only the output, but requires attention to the human-AI interaction and community context.
• To highlight the role of online communities in shaping and supporting novel creative practices such as prompt engineering and curation.
• To advocate for a shift in evaluation frameworks toward including process, person, and press (environment) in assessing creativity in AI-generated art.
• To identify opportunities for Human-Computer Interaction (HCI) research in designing co-creative systems that support prompt engineering and community collaboration.

Proposed method

• Adopted Rhodes’ '4 P' model of creativity—product, person, process, and press—as a conceptual framework to analyze human creativity in text-to-image generation.
• Analyzed case studies of text-to-image art creation, focusing on prompt engineering as a core creative practice that involves iterative refinement and experimentation.
• Examined the role of online communities (e.g., Midjourney, Colab, social media) as critical environments (press) that enable knowledge sharing, feedback loops, and collective learning.
• Explored image-level and portfolio-level curation as creative acts that reflect the creator’s intent, aesthetic judgment, and evolving style.
• Drew parallels with natural language prompting in large language models (e.g., GPT-3) to inform design principles for prompt engineering in text-to-image systems.
• Proposed that future HCI research should focus on designing user interfaces and tools that support co-creative workflows, such as MindsEye or community-driven prompt templates.

Experimental results

Research questions

• RQ1 How does the product-centered view of creativity fail to capture the full extent of human creativity in text-to-image generation?
• RQ2 What role do online communities play in shaping the creative process and practices such as prompt engineering in text-to-image art?
• RQ3 In what ways does the interaction between humans and AI systems constitute a form of co-creation that goes beyond the final image output?
• RQ4 How can the '4 P' model (product, person, process, press) be applied to better understand and evaluate creativity in AI-generated art?
• RQ5 What opportunities exist for HCI research in designing systems that support and enhance prompt engineering and collaborative creativity in text-to-image generation?

Key findings

• The product-centered definition of creativity—requiring originality and effectiveness of the final image—fails to account for the creative labor involved in prompt engineering and iterative refinement.
• Creativity in text-to-image generation is not solely in the image output but is deeply embedded in the process of crafting and refining prompts through trial, error, and community feedback.
• Online communities function as essential 'press' elements in the 4 P model, providing shared resources, templates, and collaborative learning environments that shape creative practice.
• Image-level and portfolio-level curation are significant creative acts that reflect the artist’s aesthetic judgment and evolving identity, extending creativity beyond prompt creation.
• Prompt engineering has emerged as a distinct creative craft practiced within online communities, analogous to traditional artistic techniques but adapted to natural language interaction with AI.
• The rise of text-to-image generation signals a potential societal shift in creative production, comparable to the historical impact of photography, with implications for how we define authorship, skill, and creativity in the digital age.