From Digital Prompts to Plastic Memories: The Art and Struggle of AI-Generated Action Figures
Last Revised: June 22nd, 2025
The intersection of artificial intelligence and creative expression has moved beyond simple text generation into the realm of bespoke, physical-style artifacts. For many, AI is a tool for productivity; for Mike, the creator behind the Data Viz Blog, it has become a medium for crafting digital "action figures"—nostalgic, collectible-style representations of friends, family, and music legends. This creative journey, rooted in the memories of the March 2025 Flower Power Cruise, highlights both the remarkable potential of generative AI and the persistent, often humorous challenges posed by model "hallucinations."
The Genesis of an AI Hobby
The project began as a way to commemorate the connections made during a seven-day voyage from Fort Lauderdale to stops in Falmouth, Jamaica, and the Cayman Islands. The Flower Power Cruise, a celebration of 1960s and 70s culture, featured iconic acts such as The Cowsills, Jay and the Americans, Ron Dante, and Jefferson Starship.
Inspired by the camaraderie of the trip, Mike began using ChatGPT to translate personal photographs and specific prompts into "carded" action figures—a style reminiscent of vintage toy packaging. What started as a casual experiment quickly evolved into an intensive project that would span dozens of subjects, ranging from concert performers to coworkers and grandchildren.
The Reality of AI Hallucinations
While the results are often impressive, the process is far from seamless. The AI industry frequently grapples with "hallucinations"—instances where the model generates incorrect, illogical, or unsolicited information based on its training data. Despite advancements, recent tests indicate that models like OpenAI’s o3 and o4-mini continue to exhibit hallucination rates of 30-50%.
For the creator, this means constant vigilance. “If you see a word misspelled, or multiple Groovy Bears or Milk Shakes appearing where they shouldn’t, AI is hallucinating, and it’s not me,” Mike notes. These errors serve as a reminder that even advanced large language models struggle with spatial reasoning, consistency, and the specific nuance of a user’s intent.
A Chronology of Creation: Trials and Errors
The Martha Project: A Lesson in Iteration
The creation of an action figure for a friend named Martha serves as a primary case study in the unpredictability of AI. Initial attempts to place Martha in a white shirt and pink poodle skirt resulted in the AI repeatedly injecting unrelated elements, such as multiple "Groovy Bears" and chocolate milkshakes.
Subsequent iterations brought in "The Archies," likely due to the AI’s contextual knowledge that the band, led by Ron Dante, was associated with the aesthetic of "Pops Soda Shoppe." However, the AI eventually pivoted to a Grease-themed aesthetic without instruction. Through extensive trial and error—and a refusal to accept the initial, less-than-accurate renderings—a final, proportional, and accurate version was eventually achieved, proving that "prompt engineering" is often an exercise in persistence.
Capturing the Concert Experience: Peter Noone
The action figure for Peter Noone of Herman’s Hermits represents the pinnacle of integrating real-world artifacts into digital design. During the cruise, Noone famously engaged with the audience, tossing a fan’s purse and badge onto the stage and playfully teasing a man wearing gold sparkle shoes.
By utilizing headshots of Noone and photographs of the specific items—the purse, the badge, and the shoes—Mike was able to create a unique, artifact-rich action figure. This project demonstrated that when provided with specific, high-quality visual data, the AI can successfully synthesize disparate elements into a cohesive and meaningful narrative.
The Evolution of Style: From Realistic to "Toon"
A recurring observation in this project is the degradation of visual quality during excessive regeneration. When the AI is asked to refine an image too many times, it often shifts from a realistic, detailed aesthetic to a cartoonish, distorted style. The creator likens this to the "Toons" from the film Who Framed Roger Rabbit. This phenomenon highlights a limitation in current generative models: they appear to lose "grounding" in their original instructions the more they iterate, drifting toward more generic, high-contrast caricatures.
Supporting Data: The Technical Challenges
The project also served as a testing ground for various personas and contexts. Whether it was the "Gronk and Brandon" figure for a corporate presentation or the "Purge-themed" variant created by coworker Eric Carlson, the results varied wildly based on the specificity of the prompt.
One notable challenge occurred when the AI was tasked with creating an action figure of Jar Jar Binks. The model’s strict adherence to intellectual property (IP) policies meant the request was repeatedly blocked. The compromise—a generic "space alien" figure with green tokens—underscores the guardrails placed on modern AI, which prioritize copyright protection even in experimental, non-commercial settings.
Implications for the Future of Digital Art
The broader implications of this work are significant for both hobbyists and AI developers.
- The Human-AI Partnership: The creator’s role has shifted from "artist" to "editor and curator." The AI provides the raw output, but the human remains the final arbiter of quality, accuracy, and emotional resonance.
- Memory as Data: The project proves that AI can act as a powerful tool for memory preservation. By feeding photos of grandchildren, cruise performances, and personal items into the model, the creator is essentially building a digital archive of their life experiences.
- The Persistence of Error: The fact that the AI repeatedly "fixed" a missing baby tooth on the creator’s grandson, or added unrequested items like extra Jeeps or robotic faces, indicates that the models are not yet capable of understanding the intent behind a "flaw" (such as a missing tooth).
Official Perspectives and Reflections
In reflecting on the project, the creator notes a sense of humor and patience is essential. When the AI hallucinates, it provides a glimpse into the "mind" of the machine—a space where reality is fluid, and logic is often sacrificed for the sake of pattern matching.
Despite the technical frustrations, the project has been a success in terms of community building. Whether it is Dotti, the Vice President who took the addition of a "weathered cowboy hat" in stride, or the coin-collecting friends who identified the "fakeness" of the AI-generated currency, these figures have sparked conversations and fostered a sense of shared creativity.
Conclusion: The Work Continues
As of June 22nd, 2025, the project remains an active, evolving endeavor. The creator continues to refine techniques, moving from early "Krypto" figures to more complex two-pack designs. The process of searching for photos, crafting the perfect prompt, and managing the inevitable hallucinations has become a hobby in itself.
For those looking to explore similar projects, the lesson is clear: the AI is a collaborator, not a creator. It requires specific, grounded instructions and a steady hand to guide it. As models improve, the "hallucinations" may decrease, but for now, the imperfections remain part of the charm. The project stands as a testament to how technology can be used to bridge the gap between digital potential and human nostalgia, one action figure at a time.
