| --- |
| license: other |
| license_name: multiple-permissive |
| license_link: LICENSE.md |
| library_name: onnx |
| tags: |
| - onnx |
| - onnxruntime-web |
| - background-removal |
| - image-segmentation |
| - segment-anything |
| - super-resolution |
| - image-inpainting |
| - object-detection |
| - text-detection |
| - depth-estimation |
| - image-deblurring |
| - webgpu |
| - wasm |
| --- |
| |
| # Pixelity — models |
|
|
| ONNX models (and the ONNX Runtime Web WebAssembly binary) used by **Pixelity**, a web image editor that runs entirely in the user's browser: background removal, AI object selection, object and text detection, object removal, portrait-style background blur, deblurring and image upscaling. The browser downloads each file on demand when the feature is first used; images never leave the user's device. |
|
|
| This repository **redistributes unmodified third-party files** (except where noted as a format conversion). All credit goes to the original authors. Each file keeps its original license; full texts are in [`licenses/`](licenses/). |
|
|
| ## Files |
|
|
| | File | Use in Pixelity | Origin | License | Size | SHA-256 | |
| |---|---|---|---|---|---| |
| | `models/u2netp.onnx` | Background removal (fast) | [U²-Net](https://github.com/xuebinqin/U-2-Net) (Xuebin Qin et al.); ONNX file from [rembg v0.0.0](https://github.com/danielgatis/rembg/releases/tag/v0.0.0) (Daniel Gatis) | Apache-2.0 (weights) · MIT (rembg) | 4.57 MB | `309c8469258dda742793dce0ebea8e6dd393174f89934733ecc8b14c76f4ddd8` | |
| | `models/birefnet_lite_fp16.onnx` | Background removal (high quality, GPU with fp16) | [BiRefNet](https://github.com/ZhengPeng7/BiRefNet) (Peng Zheng et al.); [onnx-community/BiRefNet_lite-ONNX](https://huggingface.co/onnx-community/BiRefNet_lite-ONNX/tree/de15b22ba131738a16dff04aab8bdf8dc32e3ac1) `onnx/model_fp16.onnx` | MIT | 114.5 MB | `d39b897ceb16ae654c1731f3dba0cf9b368d9cae74b5a57459b455cc8bfec402` | |
| | `models/birefnet_lite_fp32.onnx` | Background removal (high quality, GPU without fp16) | Same repository, `onnx/model.onnx` | MIT | 224.0 MB | `5600024376f572a557870a5eb0afb1e5961636bef4e1e22132025467d0f03333` | |
| | `models/slimsam77_vision_encoder_quantized.onnx` | AI selection (CPU), image encoder | [SlimSAM](https://github.com/czg1225/SlimSAM) (Zigeng Chen et al.), based on [Segment Anything](https://github.com/facebookresearch/segment-anything) (Meta); [Xenova/slimsam-77-uniform](https://huggingface.co/Xenova/slimsam-77-uniform/tree/5850ab45f587c112167512ffef949107115e26a0) | Apache-2.0 | 8.88 MB | `cce23c7b2e5d4f330932738fb67ba518e04b0d99ccdd1cccd22a7da4e01f2971` | |
| | `models/slimsam77_prompt_encoder_mask_decoder_quantized.onnx` | AI selection (CPU), mask decoder | Same repository | Apache-2.0 | 4.90 MB | `cb90b279f549d2cab7fd6e20c38522438c65d84bdcca3d2a764cff7d857fdce2` | |
| | `models/slimsam77_vision_encoder.onnx` | AI selection (GPU), image encoder | Same repository | Apache-2.0 | 23.28 MB | `9f8433273a6750b587779baa0cf5508111001bf7e7acfcf585d370139fd366d0` | |
| | `models/slimsam77_prompt_encoder_mask_decoder.onnx` | AI selection (GPU), mask decoder | Same repository | Apache-2.0 | 16.56 MB | `f4514391764fbd56e08e119060d874ecd7d52994bfb1968af159e12d4943b5bb` | |
| | `models/realesr_general_x4v3.onnx` | Image upscaling ×2 / ×4 | [Real-ESRGAN](https://github.com/xinntao/Real-ESRGAN) (Xintao Wang et al.), official weights `realesr-general-x4v3.pth` ([release v0.2.5.0](https://github.com/xinntao/Real-ESRGAN/releases/tag/v0.2.5.0)). **Converted to ONNX by the Pixelity project** (format conversion only, weights unchanged; verified against PyTorch, max. error 4.2·10⁻⁶) | BSD-3-Clause | 4.87 MB | `f788518cc66e302bc9b6946016e989dd9374f1012d5b6c0c7e753904ef3f99e4` | |
| | `models/migan_pipeline_v2.onnx` | Object removal (inpainting) | [MI-GAN](https://github.com/Picsart-AI-Research/MI-GAN) (Andranik Sargsyan et al., Picsart AI Research), Places2 512 model; ONNX pipeline file from [andraniksargsyan/migan](https://huggingface.co/andraniksargsyan/migan/tree/406830d0fa60666da0071c342ad2fbc8f30c5c64) (linked as the official ONNX file in the MI-GAN README) | MIT (code and weights) | 28.08 MB | `6f1f3530a1a2324b19752018ce756088b07973cda8d7d890034ace5c8a48c40b` | |
| | `models/rfdetr_nano_q8.onnx` | Object detection (fast automatic contours) | [RF-DETR](https://github.com/roboflow/rf-detr) Nano (Roboflow), COCO; 8-bit quantized ONNX file `onnx/model_quantized.onnx` from [onnx-community/rfdetr_nano-ONNX](https://huggingface.co/onnx-community/rfdetr_nano-ONNX/tree/eae21cee0687a91bcf9fa071605c48d7705d2d91) | Apache-2.0 | 28.83 MB | `2981aa0a57781c0f1a5be171c3fded504c38fe8611326bc5e3d0f8a2e9a57085` | |
| | `models/ppocrv6_tiny_det.onnx` | Text detection (select all text in a photo) | [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) PP-OCRv6 tiny text detection model (PaddlePaddle); official ONNX file `inference.onnx` from [PaddlePaddle/PP-OCRv6_tiny_det_onnx](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx/tree/2ba1506c0380b8f0b03dd142459aac66d4421f6c) | Apache-2.0 | 1.78 MB | `193bab7a04fca699a6c82e6abb5b81bdb28177f0abd4062552b04908dafb19f8` | |
| | `models/depth_anything_v2_small_q8.onnx` | Depth estimation (portrait-style background blur) | [Depth Anything V2](https://github.com/DepthAnything/Depth-Anything-V2) **Small** (Lihe Yang et al.); 8-bit quantized ONNX file `onnx/model_quantized.onnx` from [onnx-community/depth-anything-v2-small](https://huggingface.co/onnx-community/depth-anything-v2-small/tree/4472b7362082ad9968fee890ca0f1e5aca36b93d). Only the Small model is Apache-2.0 (Base, Large and Giant are CC-BY-NC-4.0 and are **not** included) | Apache-2.0 | 27.26 MB | `fcf51f1b230362b28690bb9d1809bf0431f29cad20534e3f589bd7285547f20d` | |
| | `models/nafnet_reds64_int8.onnx` | Fix blurry / motion-blurred photos | [NAFNet](https://github.com/megvii-research/NAFNet) (Liangyu Chen et al., Megvii Research), REDS width-64 model (motion blur + JPEG artifacts); ONNX file with int8 weights `deblurring_nafnet_2025may.onnx` from [opencv/deblurring_nafnet](https://huggingface.co/opencv/deblurring_nafnet/tree/f1f255116cdb628a311d2b5749871189a4639d84) (OpenCV model zoo) | MIT | 91.74 MB | `07263f416febecce10193dd648e950b22e397cf521eedab1a114ef77b2bc9587` | |
| | `ort/ort-wasm-simd-threaded.asyncify.wasm` | Inference runtime | [ONNX Runtime Web](https://github.com/microsoft/onnxruntime) 1.30.0 (Microsoft), npm package `onnxruntime-web@1.30.0` | MIT | 26.78 MB | `39f9f0894d478800487ed9f7dbe92618498db320cf55c8e3d89adff8dce658da` | |
|
|
| ## Licenses and attribution |
|
|
| All licenses allow commercial use and redistribution provided the copyright notices and license texts are kept: |
|
|
| - **MIT** — BiRefNet © 2024 ZhengPeng · NAFNet © 2022 megvii-model (ONNX by OpenCV) · MI-GAN © 2024 Picsart AI Research (PAIR) · rembg © 2020 Daniel Gatis · ONNX Runtime © Microsoft Corporation. |
| - **Apache-2.0** — U²-Net © Xuebin Qin et al. · SlimSAM © Zigeng Chen et al. · Segment Anything © Meta Platforms, Inc. and affiliates · RF-DETR © Roboflow, Inc. · PaddleOCR © PaddlePaddle Authors · Depth Anything V2 (Small) © Lihe Yang et al. The files are redistributed **without modification**. |
| - **BSD-3-Clause** — Real-ESRGAN © 2021 Xintao Wang. The authors' names are not used to endorse or promote Pixelity. |
|
|
| ONNX Runtime third-party notices: [`licenses/onnxruntime-ThirdPartyNotices.txt`](licenses/onnxruntime-ThirdPartyNotices.txt). |
|
|
| The files are provided **"as is", without warranty of any kind**, as stated in each license. |
|
|
| ## Citations |
|
|
| - Qin et al., *U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection*, Pattern Recognition 2020. |
| - Zheng et al., *Bilateral Reference for High-Resolution Dichotomous Image Segmentation*, CAAI AIR 2024. |
| - Chen et al., *SlimSAM: 0.1% Data Makes Segment Anything Slim*, NeurIPS 2024. |
| - Kirillov et al., *Segment Anything*, ICCV 2023. |
| - Robinson et al., *RF-DETR: Neural Architecture Search for Real-Time Detection Transformers*, ICLR 2026. |
| - Cui et al., *PaddleOCR 3.0 Technical Report*, arXiv:2507.05595, 2025. |
| - Yang et al., *Depth Anything V2*, NeurIPS 2024. |
| - Sargsyan et al., *MI-GAN: A Simple Baseline for Image Inpainting on Mobile Devices*, ICCV 2023. |
| - Chen et al., *Simple Baselines for Image Restoration*, ECCV 2022. |
| - Wang et al., *Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data*, ICCVW 2021. |
|
|
