** Adrianne's original post can be found here: https://blogs.ubc.ca/akraft/2025/11/16/text-to-image/
For this linking assignment, I reflected on Adrianne’s post and their “Studio Ghibli-style” text to image prompt. Adrianne used the prompt in Microsoft Copilot: “Create an image of an inclusive makerspace in studio Ghibli style.”
The resulting image was surprisingly impressive. The art style immediately evoked Studio Ghibli (warm lighting, whimsical style, and soft lines). Even without Totoro in the scene, the resemblance was unmistakable.
Adrianne’s analysis of the image was insightful and raised important points about the limitations of generative AI. Although Copilot generated this image that it determined was “inclusive”, they mentioned that Copilot doesn’t actually understand what inclusive means. Instead, it simply created an approximation based on what inclusivity would have been described as in the algorithm that it was trained on.
Adrianne captured this well, stating: “This really showed me how the AI pulls from its internal patterns and associations rather than my actual intentions.”
Reading their reflection made me curious about what Adrianne’s specific intentions were for “inclusivity” and I would have loved to see them push Copilot further by refining the prompt, requesting additional variations, or explicit prompts. Iterating on the prompt may have revealed Copilot’s limitations or alternative interpretations of what an inclusive makerspace really is.
It would be fascinating to see the training data behind Copilot’s interpretation of Studio Ghibli art, as well as what its been trained on to describe inclusivity. Access to that data would help us better understand what the model relies on, and any blind spots or biases that may emerge.
Here is Adrianne’s original post:
AI Disclaimer: ChatGPT was used to help edit my work for grammar and clarity. All final edits and ideas are my own.