Figma plugins & AIWorking prototype

Flat ads turned back into editable Figma layers.

Two prototype Figma plugins rebuild a flattened PNG or JPG as live text, shapes and cut-out images, so designers stop redrawing an ad just to change a headline.

3layer detail levels, copy only to atomic
16test designs in the accuracy check
A Figma plugin panel next to a flat ad and the same ad rebuilt as editable layers
Rebuilt interface · illustrative data

The business problem

Designers often receive an ad, or generate one with AI, as a single flat image. Changing a headline, resizing for another placement or reusing the product card means redrawing it from scratch in Figma. That rework is slow and repetitive.

What changed

Yash built two plugins. LayerLift uploads an image to a small backend. AI models find the text, shapes and logos, redraw anything hidden behind another element, and the plugin builds native Figma layers. Designers choose how far to split it: copy only, components, or every element on its own.

Relayer is the earlier, hands-on version. It rebuilds a frame from a layer description file plus a cleaned background, with the analysis done outside Figma. It was tested on a carousel panel and produced 21 layers.

Before & after

AreaBeforeAfter
A flat ad needs a new headlineRedraw the ad from scratchEdit live text on a rebuilt layer
Hidden parts of a designLost behind the layer on topRedrawn so each layer is whole
Checking the resultEyeball itThe rebuild is rendered and compared with the original

The business value

LayerLift takes roughly one to two minutes per image, according to the install notes handed to the team. It scored about 81% layer recall and 87% precision on 16 test designs in an August 2026 run, with 76% of text matched exactly.

How to read the return

This is a prototype. No time saved or adoption figures were measured. Font family matched only 43% of the time in that run, so designers still swap in brand fonts. The scores come from a single non-deterministic run and later changes were not re-scored.

Why the system is dependable

Every model response is validated before it reaches Figma, and malformed layers are dropped rather than passed on. The rebuilt design is rendered back to a flat image and compared with the source, then repaired where it falls short. An accuracy harness scores the pipeline against real layered designs, so changes can be judged by numbers. The plugin works without any model through an offline mock backend.

A closer look at the build

The Relayer plugin with a layer description file and a cleaned background beside a rebuilt frameRelayer: layer description in, editable frame out (full size, opens in a new tab)

Mock-ups use a fictional ad and brand. Both plugins are development builds loaded from a manifest, not published to the Figma community.

Could this work for your team?

If your team keeps redrawing flattened or AI-generated creative, a layering step can hand designers editable files instead of pictures.

Talk through a similar project
Keep exploring

More case studies

All case studies