Figma's New AI Tool Faces Similarity Issue, Temporarily Pulled Offline

TapTechNews July 3rd news, Figma has recently launched a new AI tool named "Make Designs", which allows users to quickly generate application prototypes. However, as many designs generated by this tool are strikingly similar to Apple's iOS Weather App, Figma had to temporarily take it offline. Figma's CEO Dylan Field explained this move on social platform X in the early hours of Tuesday, admitting that he made a mistake in setting the product release deadline and defending the company's approach to developing the AI tool.

TapTechNews noted that Andy Allen, the CEO of NotBoringSoftware, a social platform X user, demonstrated through comparison how Figma's "Make Designs" tool generated designs almost identical to Apple's Weather App. Allen wrote, "For any designers using the new 'Make Designs' feature, I suggest you thoroughly check existing applications or significantly modify the generated results to avoid getting into legal disputes unknowingly."

Figmas New AI Tool Faces Similarity Issue, Temporarily Pulled Offline_0

When TheVerge interviewed Figma's Chief Technology Officer Kris Rasmussen on Tuesday, it directly asked whether "Make Designs" used Apple's App designs as training data, Rasmussen's answer was that he couldn't be sure, and Figma wasn't responsible for training any of the AI models it used.

Rasmussen said, "The generative AI features haven't gone through any training. These features are supported by off-the-shelf models and our commissioned custom-specific design systems, and that seems to be the root cause of the problem."

This is largely consistent with what he said in his reply to a user on social platform X on Monday when the user questioned whether "Make Designs" used existing Apps as training data. Rasmussen replied, "As we shared when we launched Figma AI last week, this feature or any of our generative features have no training stage. We are investigating to what extent the similarities are due to the third-party models we use or due to the design systems used for commissioned models and will make improvements when necessary."

Field said in his post that the "Make Designs" feature "doesn't use Figma content, community files, or App designs as training data", and pointed out that "the accusation about data training in this tweet is wrong." He admitted that there was a problem with the company's approach, that is, "too little variability".

According to Rasmussen, the AI models that provide core support for the "Make Designs" feature are OpenAI's GPT-4o and Amazon's Titan image generator G1. If Figma indeed doesn't train its AI tools but they can still generate designs similar to Apple's Apps, then this may indicate that OpenAI or Amazon's models use Apple's designs as training data. OpenAI and Amazon didn't immediately respond to requests for comment.

Rasmussen argued that Figma wanted to wait until it disclosed its AI training policy to users (the company also launched this policy last week) before starting any training for improvements to the features. As part of this policy, users can decide whether to allow their content to be used for Figma's training, with a deadline of August 15th. (By default, entry-level and professional plan users choose to participate, while organization and enterprise plan users choose to opt out.)

Rasmussen also pointed out that the "Make Designs" feature is still in the testing stage, "The test version is not perfect by definition. But for sure, as Dylan said in the tweet, we just didn't spot this problem, and we should have.&q uot;

Rasmussen said Figma will "soon" re-enable the "Make Designs" feature, and other FigmaAI features will also remain in the testing stage.

Figma isn't the first company to come under review for its approach to introducing AI into its creative tools. Previously, Adobe had to clarify that it wouldn't use users' works to train its AI after strong opposition to a change in its service terms. Meta also had to change its AI tags when photographers complained that its old tags were wrongly applied to real photos.

Likes