AI model & licenses

Creatafy Workflow Background Remover uses a local ONNX inference engine and an IS-Net segmentation model. No photo data is sent to a background-removal service.

IS-Net / DIS

Highly Accurate Dichotomous Image Segmentation, Xuebin Qin, Hang Dai, Xiaobin Hu, Deng-Ping Fan, Ling Shao and Luc Van Gool. ECCV 2022.

Original project and model documentation ↗ · Upstream Apache 2.0 license

The quantized ONNX model is distributed by IMG.LY in background-removal-data 1.7.0. Its original model chunks are included unchanged and verified by SHA-256 at packaging time.

Model distribution manifest ↗ · Distributor third-party license notices

ONNX Runtime Web 1.21.0

Copyright Microsoft Corporation. Used under the MIT license.

ONNX Runtime license · Project source ↗

Processing

The page loads a quantized IS-Net model in a web worker, normalizes RGB input, predicts a foreground mask, and combines it with any existing image transparency. The website's background-removal integration is implemented directly with ONNX Runtime; it does not bundle the IMG.LY background-removal JavaScript library.