Skip to main content

Regression in Figma Make

  • June 4, 2026
  • 3 replies
  • 34 views

Sophie_Teare

I last used Figma make a month ago, and whatever change was made to the default LLM made it unusable. The default selection picks an LLM that can’t produce even the most basic protoype and is unable to follow instructions. It was a powerful tool that I highly recommended but I’d be embarrassed if someone tried to use it with the current default LLM. A lower token cost doesn’t matter when the tokens are wasted by failed outputs. 

3 replies

Celyn_L
Figmate
  • Figmate
  • June 5, 2026

Hi ​@Sophie_Teare, Celyn from Figma Community Support here.

Thank you for taking the time to share this feedback. 

To help us better understand the difference you're seeing, could you share a few examples of prompts or workflows that previously worked well but are now producing poorer results, along with any screenshots of the outputs you're seeing today?

Appreciate you taking the time to share this feedback, and I'll make sure to pass it along to the team once I have a bit more context.

Thank you!


Sophie_Teare
  • Author
  • New Member
  • June 8, 2026

Typically my workflow starts with prompting with a screenshot of my app, asking the AI to mockup the area of the app. In previous prototypes I’ve built this was a non-issue, and the AI did a great job at recreating the base app. Currently, it’s not capable of getting past this first step. I’ve prompted multiple specific fixes (Asked it to update the spelling mistakes, change the layout, fix the padding) and it struggles to even create something with that level of detail provided from the prompt, when previous a simple screenshot and a ‘Recreate this image as a base for our prototype’ would have accurate results. 

I tried about 3 different LLM choices, and with none of them able to resolve the padding/spacing issue here despite detailed instructions to do so, I’ve switched back to Claude Design. Which created it correctly from the first prompt/screenshot. 


Celyn_L
Figmate
  • Figmate
  • June 10, 2026

Hi ​@Sophie_Teare, thank you for taking the time to share such detailed feedback. The examples and comparison you've provided are really helpful in understanding the difference you're seeing in your workflow.

I've gone ahead and passed this feedback along to the team, particularly around the changes you've noticed in screenshot-to-prototype generation quality and the challenges you're encountering when trying to apply detailed refinements such as layout, padding, and spacing adjustments.

If you're comfortable sharing, it would also be really helpful if you could provide a link to an affected file so we can take a closer look and give the team additional context. Please invite support-share@figma.com with "Can View" access. This won't impact your billing and will allow us to reproduce the behavior and investigate further with Engineering.

Thank you again for taking the time to document your experience and share such thoughtful feedback 🙏