Antoni Rodriguez-Fornells, 2024-2025 FIAS FP x Iméra fellow, is co-author of the study “Stable Diffusion Models Reveal a Persisting Human–AI Gap in Visual Creativity”, which was recently published in Advanced Science in March 2026.
Testing the creative imagination of a generative AI
This research, focused on visual creativity and imagination, saw an image-generation AI model, specifically Stable Diffusion, undergoing a creative imagination task, alongside two groups of human participants: visual artists and general population.
To reproduce a visually-comparable drawing style, the generative model was trained using the human participants’ creative productions, and was later tested under two conditions: with and without human guidance, by employing elaborate, concrete prompts in the former case, and more basic ones in the latter. This manipulation aimed at identifying the potential modulatory effects of human input on the model’s successful generation of creative productions.
The study’s findings revealed that generative AI models performed poorly when it came to producing creative images.Visual artists were rated as most creative, followed by general population and the human-guided AI model, and lastly, in a strong disadvantage, the unguided AI model
Findings that contradict the current scientific consensus
Since the release of LLMs and image-generation models (e.g. Midjourney, DALL-E, Stable Diffusion) to the public, in the last few years of the AI boom, the impressive speed, detail and accuracy of generated productions, both textual and visual, have given rise to increasingly dystopic views around AI models, incredibly successful at simulating human artistic productions.
In contrast to early cognitive science research on AI models, which often reported seemingly highly intelligent and coherent results from a variety of psychometric tests, these findings go against the grain of the current consensus, which regards AI as an autonomous creative agent. By showing the fundamental need for human intervention at multiple steps of AI models’ creative process, the present research findings are ever so relevant in the problematization of the real capabilities of generative models, as they highlight the need to draw a strong distinction between the potential of reproducing convincing creative productions, and the presence of an actual creative process taking place behind the scenes of said productions.
At present, while generative AI models may indeed appear creative, when deconstructing their imaginative process, their lack of true, autonomous creative abilities becomes apparent, AI creativity thus remaining a myth.
The project was developed by an international team of researchers, formed by members of the Computer Vision Center (UAB), IDIBELL-University of Barcelona (Cognition and Brain Plasticity Unit and the Center for Language and Computation) and the Vienna Cognitive Science Hub.
Antoni Rodriguez-Fornells was able to work on this article during his residency at Iméra from September 2024 to July 2025. The residency was made possible by the FIAS fellowship Programme, supported by the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement n°945408.
