Background removal will not start: file checks or model loading?
A file rejection, an unreadable image and a model download failure need different fixes. We tested the actual file-validation function at the size boundary. The table is a validation test, not a claim about cutout quality.
Input and result
image/gif | 1,000 bytes | imageType |
image/png | 10,485,760 bytes | type/size OK |
image/png | 10,485,761 bytes | imageSize |
What the result means
An image/gif of 1,000 bytes returns imageType even though it is small. A PNG at 10,485,760 bytes passes the type/size check; one byte more returns imageSize. Passing this check does not prove that the image can be decoded. The browser then opens it and downloads the background model before processing. Model-download and processing failures have separate messages.
What to do next
1. Save a genuine JPG, PNG or WebP; renaming an extension does not convert a file. 2. Keep it at or below 10 MiB (10,485,760 bytes). 3. If it cannot open, re-export the original image. 4. For model loading, check connectivity and retry. 5. For processing failure, try a smaller image or another browser. Inspect any successful cutout against both light and dark backgrounds before publishing.
Use the toolLimits and test conditions
Metadata checks were executed locally on 2026-10-28. This test did not process a portrait or measure speed, memory or segmentation accuracy. The current implementation reduces input to a maximum 384-pixel edge before preprocessing and enlarges the output mask; original-size output does not guarantee original edge detail. Images are processed locally, but first model loading uses a network request.
workflow-results.jsonReferences
Apps24 checks examples against the implementation and regression tests. Browser experiments state their conditions; they do not certify every device. Report an error with the page URL, input, expected result and browser. Never send passwords, private documents or personal data.