IntoLayers

IntoLayers peels a flattened JPG/PNG back into a layer stack: element layers are redrawn and auto-named by a third-party commercial multimodal layer-decomposition model, the background is generatively repainted, and the result is exported as a layered PSD that keeps layer names and stacking order, per-layer PNGs, or SVG. It is built for revision work where the source file is lost and the client only has one flattened image to hand.

Core capabilities

• Layer decomposition: handled by a third-party commercial multimodal layer-decomposition model; a single image can output 2/3/5/7/9 layers or Auto (up to 17 layers), and element layers are redrawn rather than cropped.
• Layered PSD: layer names, positions and stacking order are preserved, and element layers are smart objects that can be double-clicked back to their original size; there are no layer masks, though, and text becomes pixels.
• Per-layer PNG plus layers.json: each layer ships with a coordinate and stacking-order manifest, so it can be reassembled precisely in its original position.
• Browser-side packaging: PSD and ZIP files are generated locally by ag-psd and JSZip, so images are never uploaded a second time.
• Free SVG tracing: the Cricut edition runs color quantization and WASM tracing in the browser; it is a purely local algorithm rather than AI, and it consumes no credits.
• Input and billing: one image at a time, PNG/JPEG only, ≤10MB, 262,144–36,000,000 pixels; credits are charged by the actual layer count and refunded on failure.

Where it fits Revising a design when the client supplies only one flattened image, correcting a word or two in an AI-generated picture, splitting artwork into parts for parallax animation, or tracing cut lines for Cricut are all its home ground. Batch processing, WebP/PDF input and PSDs whose text layers must stay editable, however, are beyond it.

Distinctive advantages Most comparable tools only post result images; IntoLayers publishes the method and error table behind 15 quantified self-tests, and states plainly that the background is repainted, that occluded areas are guesses, and that 2K is merely bigger rather than clearer. Billing by layer also tracks real cost more closely than billing per run.

Editorial assessment A rare product that spells out both the provenance of its model and its limits: AI is used only for decomposition and background repainting, while packaging and vectorization are labeled as local computation. It suits designers, e-commerce retouchers and animators in a hurry to turn an old image into an editable project; anyone who needs pixel-level fidelity or genuine text layers receives an approximation — the average color difference when composited over the original is 6.88/255, and occluded content may be fabricated. A 6-layer 1K job has a median of about 82 seconds, and complex cartoon characters can fail outright at 7/9 layers without any charge. Verifying with the free first job before subscribing is the safer route.