AI image detector (browser build)
Unmodified 4-bit ONNX file model_q4.onnx (67,384,061 bytes) published by WaveSpeed at wavespeed/public-misc-models revision 3cca40ed08e1e5a86007a1fd7ebce159db8df093, path ai-image-detector/model_q4.onnx.
That file is a modified export of Ateeqq/ai-vs-human-image-detector (Apache-2.0, see LICENSE): converted to ONNX and MatMul weights quantized to 4-bit (MatMulNBits, block size 16). The fine-tune is based on google/siglip2-base-patch16-224, also Apache-2.0.
Input pixel_values float32 [1,3,224,224], RGB resized to 224×224 (bilinear, no crop), scaled to x/127.5-1. Output logits [1,2]. Softmax index 0 is AI-generated, index 1 is human. The score is an estimate, not proof.
Xet Storage Details
- Size:
- 902 Bytes
- Xet hash:
- 08413290ec7d1a6c834facc72048bdc9ce229333bda0ade51bb1d904ee641696
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.