ParamExplorer: A framework for exploring parameters in generative art

TL;DR

Generative art presents a challenging, high-dimensional parameter search problem where aesthetically compelling results are rare. The authors introduce ParamExplorer, an RL-inspired interactive framework that integrates with p5.js and supports both human-in-the-loop and automated feedback, accompanied by multiple exploration agents to study search strategies. Their contributions include a modular framework with plug-and-play agents, a client–server architecture, and a tinydb-based history gallery, validated on two generative systems (Superformula and Suburbia) to demonstrate diverse, accelerated discovery. Results indicate that open-ended and seeded exploration yield richer, more varied outputs than random baselines, providing a practical tool for artists to systematically explore creative spaces. The work enables diverse, trackable exploration of generative-art parameter spaces, with potential to broaden adoption of computational creativity tools.

Abstract

Generative art systems often involve high-dimensional and complex parameter spaces in which aesthetically compelling outputs occupy only small, fragmented regions. Because of this combinatorial explosion, artists typically rely on extensive manual trial-and-error, leaving many potentially interesting configurations undiscovered. In this work we make two contributions. First, we introduce ParamExplorer, an interactive and modular framework inspired by reinforcement learning that helps the exploration of parameter spaces in generative art algorithms, guided by human-in-the-loop or even automated feedback. The framework also integrates seamlessly with existing p5.js projects. Second, within this framework we implement and evaluate several exploration strategies, referred to as agents.

Figures (3)

• Figure 1: Interaction loop between the user and the agent. The green box represents the traditional manual feedback loop between the user and the generative system. The agent extends this loop and gives the user two additional options: (1) provide feedback to the agent, which uses the score to update its internal state and propose new parameters; and (2) refine the current drawing proposed by the agent by manually adjusting the parameters. As before, the user can also generate a new image manually. Additionally, the user may give instructions to the agent, such as moving forward or backward in its internal timeline to help guide its exploration-exploitation balance.
• Figure 2: Comparison between several agents.
• Figure 3: Evolution of the Open-ended agent on the Suburbia system after roughly 200 iterations. Each row corresponds to a different internal mode of the agent and shows the last images generated from that mode.

Theorems & Definitions (1)

• Remark 1