EraseBG vs rembg

EraseBG removes image backgrounds in a browser on the user’s device, with no usage limits listed for its free plan. It accepts JPG, PNG, WebP, and HEIC images, including batches, and provides erase and restore brushes, undo and redo, background choices, and before-and-after comparison. Optional 2× and 4× upscaling runs on the device. Exports are full-resolution PNG, WebP, or JPG, and batches can be downloaded as a ZIP. Processing uses the device’s GPU or CPU; the browser uses GPU processing automatically when WebGPU is supported. The maker says images remain in tab memory and are not collected, stored, or transmitted. The model can be cached for later offline visits. The source is MIT licensed and can be copied to a web server for self-hosting. It is available on web, iOS, Android, and as self-hosted software. Safari exports above about 16 megapixels are scaled to that limit, and iOS may reduce large photos during processing before full-resolution export. There is no API.

rembg is a free tool for removing image backgrounds. It can run as a command-line program, Python library, HTTP server, or Docker container, with API access and batch processing. The command line handles individual images, folders, watch mode, and RGB24 streams; PNG is the listed export format. Models cover general images, portraits, clothing, anime, and object segmentation, with bria-rmbg as the default local model. Edge refinement options include color decontamination, alpha matting, and ViTMatte. Documentation covers FFmpeg RGB24 output, OpenCV, PIL, NumPy, Docker, and the HTTP API. Installation supports CPU, NVIDIA CUDA GPU, and AMD ROCm backends, and requires Python versions >=3.11 and <3.14. Local ONNX models download on first use into the rembg models directory. An optional withoutBG backend uploads images to its servers, with a documented 20 MB maximum. The tool is MIT-licensed, but model weights have separate licenses; commercial RMBG-2.0 use under the BRIA license requires a paid agreement.