AI's bicycle-riding pelicans spark a creative-sameness debate
A thirteen-model visual comparison asks why a playful SVG prompt produces familiar compositions; readers argue over whether the prompt or the models deserve the blame.
A familiar bird becomes a creativity argument
The bicycle-riding pelican, a recurring informal test of language models’ SVG drawing abilities, has become a new argument about creative sameness. In an October 14 post titled The Emperor Pelican Has No Clothes, Aman Rusia presents a grid of thirteen model-labeled pelicans and asks why the results look so familiar.
The prompt is simple: “Generate an SVG of a pelican riding a bicycle.” Rusia credits Simon Willison’s original runs and says the grid uses the highest reasoning setting that returned an SVG. A second grid offers twelve contrasting scenes, including a royal parade, a storm and a lunar setting. Rusia calls the contrast a diversity-collapse problem and connects it to the authors’ UI-design creativity research.
That is a visual argument, not a controlled finding that model creativity has deteriorated over time. The displayed alternatives do not establish that each model produced them under the same prompt and settings. The experiment concerns models generating SVG code; it is not a general comparison of dedicated image-generation systems.
Readers ask what the prompt actually requests
The English Hacker News discussion splits over the premise. Some participants describe similarly repetitive stories or visual ideas. Others argue that a short, unspecified prompt reasonably invites a generic composition. One reader reports getting a more elaborate circus scene after explicitly requesting a unique variation; that remains an individual prompting account.
Participants also suggest supplying outside themes or randomized words, while disputing whether randomness alone amounts to creativity. The useful question for people making illustrations and interfaces is concrete: should a model invent an unusual creative direction without being asked, or should the brief supply that direction? The pelican grid makes that disagreement easy to inspect without turning an entertaining comparison into a universal capability verdict.