What the studio
actually does.

Resolution has different meanings

The editor retains the source pixel dimensions unless you choose an operation that changes them. Background removal uses a smaller model input and then enlarges the predicted mask to fit the original. That preserves the output canvas size, but it does not make the mask as detailed as a model that evaluates every source pixel.

The workspace rejects images above its stated bounds rather than silently reducing them. A device can run out of available memory below those limits, particularly when other applications are open. Large originals, brush undo history, several canvas layers and ZIP buffers all consume memory. The limits are a practical ceiling, not a device-capacity guarantee.

Encoding and browser support

The browser must be able to decode an input before the editor can work with it. Output support is tested separately. An unavailable encoder is disabled, and the export verifies the returned MIME type so a PNG is not mislabeled as AVIF. JPG exports composite transparent areas onto white because JPG has no alpha channel.

HEIC decoding uses native browser support only. No additional HEIC decoder is shipped. Unsupported inputs get a clear error and are not sent to a remote conversion service. Animated images become a still frame, and the app does not preserve the original metadata or color profile byte for byte.

Local processing and storage

Basic tools use canvas operations. Background removal loads pinned model weights, verifies their hash, and tries WebGPU with a WASM fallback. The model cache is best effort. Images selected on the home page temporarily pass through same-origin IndexedDB and are deleted when the studio reads them. See the privacy page for abandoned-transfer and cache behavior.

Smart product framing

Frame & download runs the selected model if needed, keeps the current refined mask when it belongs to that model, and computes the bounding rectangle of pixels whose retained alpha is greater than 24 out of 255. It includes separate retained regions such as straps and accessories. A completely empty mask produces an error instead of a blank export.

The selected preset supplies the output dimensions, background color, fill percentage and format. The subject is scaled proportionally and its bounding rectangle is centered. Fill is the larger of subject width divided by canvas width and subject height divided by canvas height, expressed as a percentage. It is not the fraction of colored pixels or an official platform acceptance measurement. For example, a 200 × 400 pixel subject at 85% fill in a 1600 × 1600 frame becomes 680 × 1360 pixels, with 460-pixel horizontal margins and 120-pixel vertical margins. This calculated example shows the method, not a measured photograph.

Amazon staff guidance specifies a white main-image background and at least 85% fill. Google recommends 75% to 90% framing. Those sources do not define a measurement algorithm for irregular silhouettes. Our preset selects 85% for those two; other presets use 80% and white as our own workspace choices. Existing size guides explain whether each canvas size is a requirement, recommendation or optional choice. The tool cannot certify a category rule, repair a clipped source, verify product condition or guarantee acceptance.

Framing exports the selected preset as JPEG. Switching the general background control returns to the original canvas; Return to original canvas also clears the preset framing. Export dimensions and the chosen fill are reported after saving. Batch exports apply the active framing separately to each subject.

Batch and print calculations

Batch mode works through files sequentially and applies the current output settings. Brush edits affect only the preview image. The ZIP cap counts encoded image bytes before the archive is constructed; memory use can be higher than that count. If a batch fails, the app reports the problem instead of claiming that an incomplete archive is complete.

Print pixels are calculated as physical inches multiplied by the chosen pixel density. Millimeters are divided by 25.4 first. The sheet is a PNG with calculated geometry; it is not a printer driver. Browser metadata can use a generic density, so choose the stated paper size and disable fit-to-page scaling when printing.