Figma AI: Practical Intelligence for the Design Workflow

Figma has introduced a suite of AI-powered features designed to address common pain points in the design process. Currently in limited beta and free through 2024, these tools aim to move beyond the hype around artificial intelligence by focusing on practical solutions for searching, creating, and organizing work.

The betas can be joined directly from the Figma editor. Users can navigate to the bottom of the screen, click the "?" icon, and select Join UI3 + AI waitlist.

A significant portion of a designer's time is spent locating existing work, whether it is a production screenshot or a specific component within a sprawling design system. To tackle this, Figma is introducing two upgraded search functions.

Visual Search allows users to find designs by uploading an image, selecting an area on the canvas, or submitting a text query. The system returns visually similar designs across all team files the user has access to, allowing relevant frames to be inserted directly into the working file. Future iterations will extend this capability to community files, which will include proper attribution for the original creators. Users will also eventually be able to search files and assets from the Figma community directly within the editor.

The existing Asset Search in the Assets panel has also been upgraded to understand the semantic meaning behind queries rather than relying solely on keyword matching. This means a search for "primary button" can surface components named something like "btn_large" in the design system, providing a more intuitive way to locate and reuse design-system components.

Streamlining Repetitive Tasks

Beyond search, a set of new tools aims to reduce the tedium of common design chores, covering everything from copywriting and prototyping to layer organization.

Text, Images, and Content

AI-powered text tools let designers translate, shorten, or rewrite copy directly on the canvas to iterate on messaging more efficiently. To replace placeholder content, new content generation tools can quickly populate mockups with realistic, contextually relevant text and visuals, creating more persuasive presentations for stakeholders. Additionally, users can remove image backgrounds without leaving Figma by using an in-canvas tool to isolate subjects.

Rapid Prototyping

Static mocks can be transformed into interactive prototypes by clicking Make Prototype. The feature appears designed to shorten the path from static design to stakeholder feedback, with previews available directly on the canvas for efficient iteration.

Automatic Layer Renaming

The Rename Layers feature addresses a mundane but constant time sink. By automatically naming layers, this helps keep files organized and developer-ready, saving what could amount to hours of monotonous work over the course of a project.

Generating First Drafts From Text

Make Designs uses text prompts in the Actions panel to generate UI layouts and component options directly in the editor. It functions as a solution to the "blank canvas problem," allowing designers to describe what they need to produce a first draft they can then refine.

The feature is currently expanding its capabilities. Over time, it aims to leverage established design systems such as Google's Material 3 kit, with the eventual goal of allowing individual organizations to generate on-brand UI based on the assets and patterns unique to their own design systems. Figma sees this as a potential tool for consistently kicking off new projects within established frameworks.

Where the new features fit in

Figma’s new AI capabilities target two broad goals: immediate, practical relief and longer-term creative enablement. Some features, such as automatic layer naming and content fill, are designed to slot into existing workflows without friction. Others, like UI and prototype generation, are meant to offer a new creative starting line by helping users reach a workable first draft faster. The common thread is an attempt to solve real pain points and clear obstacles that slow designers down.

How the models are built and trained

All of the generative features launching now are powered by third-party, out-of-the-box AI models. None of them were trained on private Figma files or customer data. The fine-tuning done for Visual and Asset Search used images of user interfaces taken only from public, free Community files.

Looking ahead, Figma sees room to accelerate work by developing new models that work more efficiently with Figma-specific concepts and tools. Making those improvements requires training models to better understand design patterns, Figma’s internal formats, and its structure through Figma content. Two points are emphasized: admins have full control over whether their team’s content is used for training, and opting into AI content training is not required to use Figma or its AI features.

Data privacy and security

The model development process is intended to protect privacy and confidential information. For all customer data, Figma states that it encrypts data at rest and in transit, uses security measures against unauthorized access, and enforces tailored permissions and user access controls. Additional steps remove identifying details from content and redact sensitive information, including from text and images, so models learn general design patterns rather than specific user content.

Content training controls and defaults

A new team-level setting lets admins control whether customer content is shared with Figma for AI training. Customer content includes file content created in or uploaded by a user, such as layer names and properties, text and images, comments, and annotations. Sharing this content is optional; the preference goes into effect on August 15, 2024. If an admin disables content training after that date, newly created content and edits will not be used for training.

Starting today, admins can set this preference directly in settings across all plans. The defaults vary by plan:

  • Starter and Professional plans are opted in by default but can opt out.
  • Organization and Enterprise plans are opted out by default, reflecting the more complex agreements and specific restrictions these customers typically have.

No content training occurs until August 15, providing a window to adjust the setting. Content generated by Figma AI is treated as customer content data, and users retain rights to outputs generated while using Figma AI.

Usage data is treated separately. It relates to how Figma is accessed and used, including technical logs, metadata, telemetry, and information about how content is used, such as access counts. It does not include the content itself and is used in an aggregated, de-identified way.

Community files and generative training

Free files in the Figma Community are available under licenses that allow transformation but also require attribution. What counts as attribution in AI output remains a debated topic. Until that is settled and communicated, generative models that output design will not be trained on Community files. To date, public, free Community files have been used only to improve search, including the new semantic and visual search features. Paid Community files will not be used for training.

Figma’s stated goal is to build AI in service of designers and product teams with a responsible, transparent approach. The long-term aim is for AI to become a genuine creative partner as it combines with the contributions of the broader design community.