Web and app design tool maker Figma has temporarily suspended its generative AI “Make Design” feature because it believes Apple’s Weather app is the be-all and end-all for mobile weather forecasting.
The ability to create app designs from AI text has an affinity for mimicking Apple’s design language, discovered after Andy Allen, founder of NotBoring Software, asked the platform to generate a “not boring weather app.”
Allen said the generated app would provide roughly parity with the Apple Weather app that ships with every iOS device. This behavior led Allen to speculate that Figma may have used existing app designs to train its service. “The Figma AI appears to be heavily trained on existing apps,” he wrote in a follow-up post.
Following this discovery, Figma CEO Dylan Fields wrote in a Xitter thread that the company would “temporarily disable the Make Design feature” until a fix could be put in place to prevent this behavior.
Fields also attempted to deny claims that the service was trained on designs from popular third-party apps: “The Make Design feature is not trained on Figma content, community files, or app designs. In other words, the accusations about data training in this tweet are false,” he wrote in response to Allen’s findings.
But Fields continued that the feature was built using off-the-shelf large-scale language models, which worked in conjunction with a “design system” commissioned by Figma. He explained that the problem lies with these “design systems,” adding that copying Apple’s weather software could have been prevented with additional quality assurance steps.
One interpretation of this is that Figma commissioned a number of designs to train its generative tool, and some of those designs were very similar to Apple’s, resulting in this feature being output.
“I hate to miss the mark, especially on something that I believe is fundamentally so important to the future of design,” Fields concluded.
talk RegistryAllen said he thought Field’s explanation made sense, but noted that he never claimed the service was trained on user or community data.
“The real problem between GenAI companies and creators is that none of the companies are revealing how these models are trained, what data they are trained on, and what rights are secured. It’s like fast food of creativity, where all the ingredients and processes are hidden,” he lamented.
Allen also made it clear that he is not opposed to generative AI in app design.
“GenAI seems generally good for creating intermediate things as reference material, but I wouldn’t use it for most finished pieces. It’ll get better over time and has a long future,” he said, adding that other AI features Figma has introduced, like auto-naming layers and localization, are great. “The design creation feature is something I probably needed to spend a bit more time on.”
While Fields claims that the “Make Design” feature wasn’t trained directly on existing apps, but rather only on blueprints that presumably closely resemble them, the service’s behavior raises questions about who is liable if a model produces something that appears to infringe copyright.
Several tech giants, including Google and Microsoft, have extended limited legal protections against copyright infringement claims to users of their AI-generated services.
Figma did not directly address this. RegistryThe company did not respond to questions about why or how the Make Design feature behaved the way it did, or what legal protections it currently has or plans to offer users in the future. Instead, the spokesperson pointed us to Fields’ Xitter post and a page detailing the company’s AI approach.
In light of Make Designs’ actions, Allen wrote a separate letter suggesting that users of the service “thoroughly check their existing apps or make significant changes to their outcomes” to avoid future legal troubles.®