Figma | GPT-5.6 Improves First-Pass Prototypes in Make

GPT-5.6 is now available in Figma Make, giving designers another model for turning prompts and existing design files into interactive prototypes. Released on July 9, 2026, the model is focused on producing stronger first results, adapting layouts across screen sizes, preserving the structure of supplied designs, and continuing when a build encounters certain errors.


GPT-5.6 model available in Figma Make for interactive and responsive prototypes

{getToc} $title={Table of Contents}

Figma Make gets a stronger starting point for interactive ideas


Moving from an interface idea to a useful prototype usually takes more than one prompt. The first result may have the right general direction but still need corrections to spacing, visual hierarchy, interactions, content, or responsive behavior before a team can evaluate it properly.


GPT-5.6 is intended to reduce some of that early correction work. The examples released with the model show complete interfaces created from relatively compact instructions, including dashboards, audio controls, product pages, navigation, populated interface states, and layouts that continue working across different screen sizes.



Existing Figma designs can become working prototypes


One of the most relevant uses for web designers is building from an existing Figma Design file. Instead of describing the entire interface again through text, a designer can provide the visual source and ask Figma Make to turn it into an interactive result.


In a published audio player example, GPT-5.6 preserved the layout, visual hierarchy, proportions, spacing, and overall styling of the original design. Playback controls and audio files also worked in the generated version. This can make the transition from a static screen to a testable experience more direct, especially when a team needs to review behavior before development is complete.


First passes include more complete interface details


GPT-5.6 can also add supporting details that make an early prototype easier to evaluate. A stock dashboard example included search, keyboard shortcuts, sample performance data, sparklines, and a dense visual system from one prompt.


Another test produced a content-heavy product page with measurements, care information, dropdowns, image controls, and a clickable menu. The layout adjusted across screen sizes without requiring an additional prompt. For template creators, this can provide a faster way to explore how a page structure behaves when it contains realistic amounts of content.


Automatic error recovery can reduce repeated prompting


A generated build does not always work correctly on its first attempt. GPT-5.6 can investigate some errors and try to repair the result instead of stopping immediately. In one example, the model identified why a build appeared blank and corrected the problem without requiring the user to diagnose it manually.


This does not guarantee that every generated prototype will work or that the resulting code is ready for publication. It can, however, reduce interruptions during early exploration by handling certain problems before the designer needs to step in.


IMPORTANT: GPT-5.6 is available through the Figma Make model selector on all plans, but the team's AI toggle must be enabled. Non-default models may behave differently, and credit consumption can vary depending on the selected model, task complexity, and reference materials. Review your credit history and test important interactions before relying on a generated build.{alertWarning}

Daisuki's Take: What This Means for Web Designers


For web designers, the main benefit is not simply generating an interface faster. A stronger first result gives the team something more complete to review, test, and discuss before spending time correcting basic layout or interaction problems.


We think the ability to build from an existing design is more valuable than generating pages from a blank prompt. It keeps the designer's hierarchy, proportions, and visual direction closer to the center of the process instead of asking the model to make every design decision independently.


The practical approach is to use GPT-5.6 for prototypes and early validation, then review the responsive layout, accessibility, generated content, interactions, and code before moving toward production. A polished first pass can save time, but it should still be treated as a starting point rather than a finished website.



Sources and Recommended Links