Affinity’s September 2026 update tackles that exact headache with an AI-powered feature officially referred to in Affinity and Canva documentation as Scene Decomposition—and surfaced inside the app interface as Decompose Layer. A hands-on evaluation by Diana Todea at How-To Geek found that the tool does a remarkably logical job of carving a flattened raster image back into workable, organized layers, though it comes with strict technical boundaries and an occasional run-in with Canva’s automated content filters.
What Scene Decomposition Actually Does
Introduced alongside Affinity’s Blend Tool, in-app scripting, and expanded RAW photography controls in the September 2026 release (which How-To Geek identifies as version 3.3), Scene Decomposition is built to turn flattened assets into separated elements ready for recompositing or removal.
When you feed a flat image into the tool, it does two distinct jobs:
- Compositional Separation: Rather than chopping an image into arbitrary geometric patches, the AI analyzes the scene's visual hierarchy. According to How-To Geek's testing across both graphic designs and photographs, the tool consistently separates backgrounds, foreground subjects, headings, and smaller text blocks into individual layers, organizing them with descriptive layer names.
- Generative Inpainting: Separating a foreground subject from a flat PNG normally leaves a gaping blank hole behind it. Affinity’s Scene Decomposition uses generative AI to predict and fill in the occluded background pixels that sat behind the extracted elements, giving you freedom to move or delete an object without exposing raw transparency.
In How-To Geek's testing, the workflow proved straightforward: users select the Canva AI mode from the Studio Switcher bar at the top of the workspace, open the flattened image, and click the Decompose Layer icon in the upper-right panel. Todea noted that the tool offered two quality settings—SeeDream Layerise (Ultra) and SeeDream Layerise 4K (Ultra), powered by ByteDance's SeeDream model—and took roughly two minutes per image regardless of visual complexity.
Section Summary: Scene Decomposition splits flat images into descriptively named layers and uses generative fill to reconstruct hidden backgrounds, taking about two minutes per image in hands-on testing.
The Reality Check: "Editable" Means Pixel Layers, Not Native Vectors
Before you delete your backup drives, it is critical to understand what "editable" means from an engineering standpoint. A flattened PNG or JPG contains only the final visible grid of pixels; all semantic metadata—such as which font family was used, where Bezier vector nodes were placed, or what non-destructive adjustment sliders were applied—is permanently gone.
As a result, Scene Decomposition outputs pixel layers, not reconstructed native design objects. Here is how that boundary plays out in practice on your Windows workstation:
| Capability | What You Can Do After Decomposition | What You Cannot Do |
|---|---|---|
| Layout & Positioning | Move, resize, rotate, reorder, group, align, or delete individual visual elements. | Recover the exact original layer hierarchy or non-destructive smart objects from the lost source file. |
| Typography | Isolate headings and text blocks onto their own transparent pixel layers to reposition, recolor via overlays, or delete them. | Highlight text with a cursor to fix atypo, swap fonts directly, or adjust kerning and leading as live text. |
| Vector Graphics | Extract logos and shapes as clean raster cutouts and apply layer effects like drop shadows, bevels, glows, or blend modes. | Grab vector nodes or paths with a pen tool to reshape geometry without raster scaling loss. |
| Complex Scenes | Effortlessly pull apart clear graphics and standard portraits. | Expect every tiny background detail in an intricate composition to get its own layer; the AI prioritizes primary foreground elements. |
Even with these raster limitations, having clean, isolated pixel layers with a reconstructed background saves massive amounts of manual masking time. If you need to update a date on an old promotional graphic, you can simply hide or delete the isolated pixel layer containing the old text and type fresh live text right over the newly filled background.
Section Summary: The tool creates separated pixel layers rather than live text or vector paths, meaning you can easily rearrange, mask, or delete items, but you must manually retype text if wording needs to change.
The Catch: Overzealous Content Filters and Subscription Tiers
While the compositional split impressed in testing, How-To Geek highlighted a frustrating friction point: the Canva AI content moderation filter. During testing, the filter occasionally blocked the decomposition of completely benign, original designs created by the reviewer. While there was no obvious rule governing what triggered a rejection, the reviewer observed that portraits of people and generic design layouts seemed more prone to false positives. Retrying the exact same file sometimes pushed it through, and users can report false positives when the filter misfires, though there is no immediate override when a project is on a tight deadline.
Beyond moderation quirks, Windows users should also note the licensing structure. According to Canva’s official documentation, the core Affinity application—including its standard photo, vector, and layout studios—is free to use with a verified Canva account. However, unlocking the Canva AI Studio inside Affinity requires a qualifying paid subscription.
While How-To Geek specifically references Canva Pro, Canva’s help documentation clarifies that Premium and Ultra AI tools are available across several paid tiers, including Canva Pro, Canva Teams, Canva Business, Canva Enterprise, and Canva Nonprofit (in Canva Education, Affinity AI tools are currently restricted to administrators and teachers). These cloud-backed features draw from a shared monthly AI usage allowance that resets on your billing date. For privacy-conscious professionals worried about cloud processing, Canva states in its ProSuite launch documentation that user files remain stored locally on the device and that customer work is not used to train its AI features.
Practical Advice for Salvaging Flattened Assets
If you find yourself staring at a flattened PNG on your PC, a disciplined triage approach will save you both time and AI allowance credits:
- Verify Your Backups and Filters First: Before reconstructing a file via AI, check your local Windows File History, OneDrive version history, or Canva’s project search. Canva notes in its support guides that its default homepage view only displays folders, designs, and brand templates; switching your search scope to All your content and clearing date or owner filters often surfaces "lost" source files.
- Use Decomposition for Salvage, Not Archival Restoration: Treat Scene Decomposition as a rapid compositing shortcut. Use it to strip out outdated elements, change aspect ratios for social media, or isolate subjects for photo manipulation.
- Rebuild Critical Brand Geometry: If a decomposed layer contains a company logo or headline that needs to be scaled up significantly for print, use the extracted pixel layer as a visual placement guide, then drop in a fresh vector or live font layer on top so your final export stays razor-sharp.
For Windows creatives exploring this workflow, you can find more tips on asset management, raster-to-vector workarounds, and graphic software configurations right here in our Design and Creative Software discussions on WindowsForum.com.