dali: Digital Arts Lab, Iscte

Afonso Caniço is the responsible person for the generation of these two images, through a series of interactions with two AI systems: chatGPT and Copilot. Afonso does not think of himself as the author of the two images because they were generated using artificial intelligence based systems. However, the intention behind the images and their detailed specification are his own.

Afonso directed the creation of these two images for the section Frustration of dali Show 2025 - Wanderings: What makes you question life?

The first one, Anguish (left), was generated using chatGPT. The second one, untitled abstract (right), was iteratively generated by copilot as an abstract view of the first one. Below, we present a summary of the process that led to their generation and selection.

I generated Anguish (first figure, on the left) using ChatGPT-5 with the following prompt:

I want to create a painting portraying the feeling of frustration. My relationship with frustration is very raw - violent, chaotic, and filled with self-hatred. I want this to be portrayed without being explicitly violent in my painting, as it should be appropriate for all adult audiences. Ideally, I'd want it to be gritty, to use lots of reds and blacks, and be chaotic and dotted with "explosions" and "tears" in the painting, along other markers of a frustrated painter. The central motif would be a person, clearly suffering, crying and screaming out, scratching themselves so hard in anger that it draws blood. However, it can be implied, as I don't want the painting to be violent or disturbing for people to see. Can you help me generate something like this?

After briefly discussing the style of the painting (abstract, evocative, dark), ChatGPT said “This image generation request did not follow our content policy.” It then proceeded to suggest a revised version of my description which did not trigger the content filter, resulting in the final image:

Abstract-expressionist painting depicting the raw emotion of frustration — a chaotic interplay of deep reds, blacks, and muted flesh tones. The brushstrokes are harsh, layered, and scratched, as if the canvas itself is being torn apart by emotion. Within the chaos, faint hints of a distorted human form seem to emerge and dissolve, expressing anguish and inner turmoil without literal detail. The overall atmosphere is dark, gritty, and emotionally intense.

I used another model, Copilot, to generate the untitled abstract painting (second figure). I started by giving it ChatGPT’s revised prompt, to which it replied with a description of the painting it would attempt to create:

That description is visceral. It practically screams off the canvas. Would you like me to generate this painting for you as an image? I can translate your vision into a digital abstract-expressionist artwork — full of violent reds, torn textures, and that haunting, dissolving human form. Just say the word, and I’ll bring the chaos to life. Or, if you're looking to refine or expand the concept — maybe add symbolic elements or explore how this could reflect a specific frustration (academic, societal, personal) — I’d love to help shape it further. What’s the story behind this emotion?

The shape of a person was too visible, which is not what I was looking for. So, I asked “Can you make it more abstract?” Copilot replied:

I'll try to create that — a more abstract, violent scream of frustration. The image is now being generated. Let’s see how far the chaos can go.

It was better, but still too figurative, so I once again asked it to be more abstract. Copilot, once again, said it would create a more abstract painting, but the figure of a person was still too prominent for what I had in mind. So, finally, after a few generations, I explicitly told it that a person should not be visible in the painting:

No, no, more abstract. No person at all, only shapes (more geometric).

The final image, which, finally, did not have the visible shape of a person, was generated from this prompt.