Affinity can split your PNG into layers. It hasn't found your missing design file.

Decompose Layer arrived with Affinity's September update; this week's hands-on coverage gives a closer look at what the feature actually leaves you able to edit. The distinction matters more than another list of new AI buttons.

Affinity's help says the tool separates a flat image into component parts, such as a subject, background and foreground. It replaces the selected layer with a group of separated layers. Duplicate the original first if you want to keep it alongside the result.

The help page explicitly calls this a generative separation. Where one object hid another, the model fills the gap with something plausible. It cannot retrieve detail that was never visible in the flattened image. A convincing result is still a new interpretation of that area.

The reported output is pixel layers. You can reposition elements, hide them, mask them or change how they overlap. You should not expect the tool to recover a font, turn lettering into live text or restore editable vector paths. How-To Geek's reviewer makes those limits clear.

For a simple social graphic whose background needs more breathing room, that may be enough. For a price list needing new figures, expect to replace the lettering yourself. For a logo whose curves must match the approved artwork, find the source asset rather than treat a generated reconstruction as the master.

Decompose Layer requires a Canva premium plan, according to Affinity's help. The core Affinity app being free does not make this feature free. If you already have access, try a duplicate of one ordinary image. If you don't, establish whether moving pixels would solve your problem before subscribing.

Use an image you are entitled to process and keep the original untouched. In Canva AI Studio, select the pixel layer, choose Decompose Layer, select an available AI Quality model and run it. Affinity also documents the menu route as Layer → Convert → Decompose Layer. Model choices can change.

Make the revision you actually needed. Move the object, remove the foreground element or open space around the heading. Then inspect the newly exposed area at the size where the image will be used. Check edges and shadows, but also check whether an invented detail changes what the picture says.

How-To Geek's reviewer reports processing taking around two minutes across the images tried, along with some content-filter refusals. That is one reviewer's experience, not a promised completion time or a measured success rate for your files. This article is based on the sources, not our own hands-on test.

If the useful change still requires rebuilding most of the composition, stop there. A tool can work as advertised and still be the wrong way to finish this particular job.

Once the revision is approved, save the working document with its layers. Keep the flattened original too, and label the new document as reconstructed rather than recovered. Someone opening it next month should not have to discover which parts were inferred by moving an object.

For an illustrative background, a generated patch may be entirely acceptable. For a product photograph where surface detail matters, it deserves much closer scrutiny. Moving an object out of the way does not turn the invented pixels underneath into evidence of what was there.

I like a tool that can save a design from a needless rebuild. I don't want its success to become permission to stop keeping the source files. Save the editable version while you still know where it is.