Not long ago, designer Norma Kamali saw a photo of actress Jessica Biel wearing one of her dresses, a body-hugging piece made of alternating bands of black stretch jersey and nude mesh. The image inspired Kamali to introduce new versions of the style for her spring collection, but while trying to create them, she felt stuck.

So she turned to artificial intelligence.

Kamali used a tool custom-built for her by the generative-AI creative studio Maison Meta to conjure numerous options to choose from, all of which, in her opinion, looked like they were designed by her. In a sense they were. Maison Meta’s AI was trained exclusively on Kamali’s past work as a way to preserve her design legacy, meaning any “new” ideas it produced reflected back her own creative history.

Kamali’s new AI-assisted dresses and the earlier styles they reference stand near the entry to a new installation by the designer in New York intended to give viewers a glimpse into how she’s using the technology. It’s called Fashion Hallucinations, referring to the term for the warped or inaccurate outputs AI can yield. In the space are eight-foot cutouts of stretched-looking AI-generated models frozen in motion, while a video with a voiceover by Kamali explaining her use of AI plays at intervals.

Nearly two years since ChatGPT introduced the general public to the concept of generative AI, sparking a race in fashion to find uses for the technology, questions about its transformative potential abound. A number of doubters believe its promises have been oversold, pointing to issues like hallucinations.

Kamali, who has now been using AI in her design for several months, is not among them. She appears as convinced of AI’s power as ever.

“For me, it’s really the beginning of a new way to express creativity,” she said.

So far Maison Meta has only trained the AI to produce a few categories of products, including dresses and swimsuits, but it’s changing her creative process in significant ways. For example, instead of trying to sketch out ideas visually, she has to instruct the AI via written prompts. (To make sure it understands them, Maison Meta had to train the AI on an extensive vocabulary of fashion terminology.) Then she essentially acts as a curator in selecting the best results.

In their raw form, those results can look slightly distorted. The video that plays shows the AI’s raw output cleaned up just one degree. The models’ limbs swirl and morph as they move around, momentarily producing oddities like a third leg that just as quickly disappears back into the body. It’s like a first draft that still needs polishing.

The AI Kamali uses is capable of refining the images until their flaws have been stripped away, but she chose to use the unrefined versions in the installation, showing how AI can be used to “evolve and enhance the imperfect images,” according to a panel in the space. Designers who’ve begun experimenting with the technology have pointed to hallucinations as a strength when using it for creative purposes. If it replicates things too perfectly, it’s no longer providing fresh ideas, which is why as the technology improves and hallucinations are reduced, it could diminish AI’s creative power. (What’s called the “temperature” of AI models can be adjusted, however, to increase or decrease the randomness of its output.)

What that might mean in the future is still unclear, as all of this is essentially uncharted territory. To even be able to work this way, Kamali needed an AI model created just for her. A number of start-ups are hurrying to build generative design platforms capable of letting users easily train models on their work and produce designs that are readily manufacturable.

How consumers might respond to a world of AI-assisted design is also to be determined. So far, brands have faced backlash online when customers discovered they were using AI. Critics have lambasted the technology’s developers for training their AI models on copyrighted material without consent and point to its significant environmental impact.

But increasingly, companies are testing out the technology behind the scenes, looking for ways to increase their efficiency or leverage their archives.

Kamali said her reason for doing the installation was to share her experience with other designers and creatives who might be nervous about AI or reluctant to try it out. She wants them to “just relax and have this extraordinary experience.”

“Out of any period of time in my life, this is the most exciting, because we are really just a step right into the future,” she said.