Aixio Layer v2 — Image-to-Layer AI Model

Flat design in. Editable PowerPoint out.

On a 150-design image-to-layer benchmark, Aixio Layer v2 recovered 8,096 valid editable layers: 1.7× the next best system and 2.1× Canva Magic Layers, in about 47 seconds per image.

Try in Aixio Studio · Watch the video · Benchmark · Samples · What you get back · Layer v1

_{Introducing Aixio Layer v2 — click to watch on YouTube}

Layer v2 returns every part of a design in the form you would have built it in:

• Editable text: copy comes back as real text boxes you can retype, with the matched font embedded in the file.
• Native shapes and lines: badges, circles, rules and dividers come back as PowerPoint shapes you can recolour.
• Clean image layers: photos, products, illustrations and lettering each come back on their own layer.
• A clean background: the page background comes back as its own layer.
• One native PowerPoint file, in about 47 seconds per image.

• Most valid editable layers overall: 8,096, against 4,697 for the next best system.
• Most valid text and image layers of any system tested.
• Best reconstruction score (LPIPS 0.1382), narrowly ahead of Canva Magic Layers (0.1414).
• GPT-6 Astra Ultra recovers slightly more valid vector layers (2,345 vs 2,233).

All four systems received the same 150 designs: posters, slides, infographics, social graphics, product pages and interfaces, both human-designed and AI-generated. For each design we counted the editable elements that came back usable: