- withoutBG Open Weights
- Open weights · commercially permissiveLicense: Apache-2.0
- withoutBG Pro Model Self-Host
- ProprietaryLicense: Commercial self-host license
withoutBG Open Weights vs Pro Model Self-Host
Quantitative comparison: 60 images scored against ground-truth mattes. Lower is better. Use the view controls on each row to switch between transparency, chroma key, and alpha matte on all outputs.
| Metric | withoutBG Open-Weight Model | Pro Model Self-Host |
|---|---|---|
| MGE (Edge Quality) | 0.04523 | 0.03692 ✓ |
| MAE (Overall Accuracy) | 0.02923 | 0.02759 ✓ |
Disclosure: withoutBG includes BiRefNet
withoutBG Open Weights and Pro Model API are routed pipelines. A classifier sends fine strands, soft detail and transparency to withoutBG's own matting model, and hard opaque objects, flat scenes and vehicles to BiRefNet (MIT license), which is part of the pipeline. On this page, 17 of 58 images went to BiRefNet. The table splits the scores by route: the first group is what withoutBG's own matting model handled.
| Images | Metric | withoutBG Open-Weight Model | Pro Model Self-Host |
|---|---|---|---|
| Routed to withoutBG matting (41) | MAE | 0.01846 | 0.01659 ✓ |
| MGE (edges) | 0.03773 | 0.02821 ✓ | |
| Routed to BiRefNet (17) | MAE | 0.01509 | 0.01421 ✓ |
| MGE (edges) | 0.04305 | 0.03760 ✓ |
Per-image averages, lower is better.



- MGE
- 0.03254
- MAE
- 0.00960
- MGE
- 0.02588✓
- MAE
- 0.00859✓



- MGE
- 0.06122
- MAE
- 0.02110
- MGE
- 0.04788✓
- MAE
- 0.02000✓
Photo by Alexander Dummer on Unsplash



- MGE
- 0.02084
- MAE
- 0.00601✓
- MGE
- 0.02005✓
- MAE
- 0.00767



- MGE
- 0.01262
- MAE
- 0.00461
- MGE
- 0.00929✓
- MAE
- 0.00370✓



- MGE
- 0.17462
- MAE
- 0.19712✓
- MGE
- 0.17212✓
- MAE
- 0.19787
Photo by Andreas Rasmussen on Unsplash



- MGE
- 0.07595
- MAE
- 0.02381
- MGE
- 0.04235✓
- MAE
- 0.00729✓



- MGE
- 0.03402
- MAE
- 0.02906
- MGE
- 0.03079✓
- MAE
- 0.02759✓



- MGE
- 0.04621
- MAE
- 0.01031
- MGE
- 0.04103✓
- MAE
- 0.00914✓



- MGE
- 0.01697
- MAE
- 0.00407
- MGE
- 0.00787✓
- MAE
- 0.00233✓



- MGE
- 0.01899✓
- MAE
- 0.00660✓
- MGE
- 0.02477
- MAE
- 0.00826



- MGE
- 0.04118
- MAE
- 0.01690
- MGE
- 0.02389✓
- MAE
- 0.01407✓



- MGE
- 0.04200
- MAE
- 0.01715✓
- MGE
- 0.03683✓
- MAE
- 0.02790



- MGE
- 0.05736
- MAE
- 0.08254✓
- MGE
- 0.05730✓
- MAE
- 0.09005
Photo by Camila Quintero Franco on Unsplash



- MGE
- 0.01469
- MAE
- 0.00695
- MGE
- 0.01250✓
- MAE
- 0.00669✓



- MGE
- 0.06638✓
- MAE
- 0.01242✓
- MGE
- 0.07267
- MAE
- 0.01349



- MGE
- 0.01093✓
- MAE
- 0.00177✓
- MGE
- 0.01582
- MAE
- 0.00245
Photo by Carlo Sierra on Unsplash



- MGE
- 0.02498
- MAE
- 0.00460
- MGE
- 0.01961✓
- MAE
- 0.00347✓



- MGE
- 0.03139
- MAE
- 0.00959
- MGE
- 0.02067✓
- MAE
- 0.00736✓



- MGE
- 0.02976
- MAE
- 0.00740
- MGE
- 0.02134✓
- MAE
- 0.00599✓
Photo by Conor Samuel on Unsplash



- MGE
- 0.07155
- MAE
- 0.01159
- MGE
- 0.02556✓
- MAE
- 0.00413✓



- MGE
- 0.03577
- MAE
- 0.00660
- MGE
- 0.01451✓
- MAE
- 0.00288✓



- MGE
- 0.01789
- MAE
- 0.00452
- MGE
- 0.01127✓
- MAE
- 0.00337✓



- MGE
- 0.03015
- MAE
- 0.01049
- MGE
- 0.02593✓
- MAE
- 0.00883✓



- MGE
- 0.05026✓
- MAE
- 0.01100✓
- MGE
- 0.05422
- MAE
- 0.01279



- MGE
- 0.01489
- MAE
- 0.00434
- MGE
- 0.01000✓
- MAE
- 0.00365✓



- MGE
- 0.01399✓
- MAE
- 0.00246✓
- MGE
- 0.01873
- MAE
- 0.00318
Photo by Lars Bo Nielsen on Unsplash



- MGE
- 0.05191
- MAE
- 0.01377✓
- MGE
- 0.04082✓
- MAE
- 0.01465
Photo by Mayer Tawfik on Unsplash



- MGE
- 0.01835
- MAE
- 0.00586
- MGE
- 0.00720✓
- MAE
- 0.00366✓



- MGE
- 0.01551
- MAE
- 0.00578
- MGE
- 0.00829✓
- MAE
- 0.00407✓



- MGE
- 0.02743
- MAE
- 0.00856
- MGE
- 0.02488✓
- MAE
- 0.00741✓
Photo by Niko Tsviliov on Unsplash



- MGE
- 0.14105
- MAE
- 0.05132
- MGE
- 0.08980✓
- MAE
- 0.04218✓
Photo by Norbert Braun on Unsplash



- MGE
- 0.01083
- MAE
- 0.00164
- MGE
- 0.00162✓
- MAE
- 0.00030✓



- MGE
- 0.05432✓
- MAE
- 0.02385
- MGE
- 0.05502
- MAE
- 0.02375✓



- MGE
- 0.03119
- MAE
- 0.02758
- MGE
- 0.02383✓
- MAE
- 0.02265✓



- MGE
- 0.02442
- MAE
- 0.01933
- MGE
- 0.02014✓
- MAE
- 0.01811✓



- MGE
- 0.02832
- MAE
- 0.01777
- MGE
- 0.02213✓
- MAE
- 0.00621✓



- MGE
- 0.37410
- MAE
- 0.49629
- MGE
- 0.36774✓
- MAE
- 0.49473✓



- MGE
- 0.02225
- MAE
- 0.00549
- MGE
- 0.02146✓
- MAE
- 0.00546✓



- MGE
- 0.01052
- MAE
- 0.00205
- MGE
- 0.00508✓
- MAE
- 0.00117✓



- MGE
- 0.02272
- MAE
- 0.00521
- MGE
- 0.00974✓
- MAE
- 0.00278✓



- MGE
- 0.03536✓
- MAE
- 0.00578✓
- MGE
- 0.04221
- MAE
- 0.00769



- MGE
- 0.02418
- MAE
- 0.02733
- MGE
- 0.02093✓
- MAE
- 0.02595✓



- MGE
- 0.02203
- MAE
- 0.00368
- MGE
- 0.01223✓
- MAE
- 0.00217✓



- MGE
- 0.01403
- MAE
- 0.00312
- MGE
- 0.00706✓
- MAE
- 0.00208✓



- MGE
- 0.02054
- MAE
- 0.00590
- MGE
- 0.01306✓
- MAE
- 0.00461✓



- MGE
- 0.00960✓
- MAE
- 0.00134✓
- MGE
- 0.01927
- MAE
- 0.00259



- MGE
- 0.02730
- MAE
- 0.00734
- MGE
- 0.02230✓
- MAE
- 0.00585✓



- MGE
- 0.02940
- MAE
- 0.01110
- MGE
- 0.01365✓
- MAE
- 0.00765✓



- MGE
- 0.18102
- MAE
- 0.03429
- MGE
- 0.13057✓
- MAE
- 0.02597✓



- MGE
- 0.07739
- MAE
- 0.09324
- MGE
- 0.07300✓
- MAE
- 0.09074✓



- MGE
- 0.13603
- MAE
- 0.02861
- MGE
- 0.12568✓
- MAE
- 0.02596✓



- MGE
- 0.02226✓
- MAE
- 0.00320✓
- MGE
- 0.02966
- MAE
- 0.00408



- MGE
- 0.03189
- MAE
- 0.00686
- MGE
- 0.00371✓
- MAE
- 0.00136✓



- MGE
- 0.01003
- MAE
- 0.01228
- MGE
- 0.00719✓
- MAE
- 0.01014✓
Photo by Vinicius Wiesehofer on Unsplash



- MGE
- 0.02822
- MAE
- 0.01994✓
- MGE
- 0.02527✓
- MAE
- 0.02075



- MGE
- 0.06085
- MAE
- 0.24387
- MGE
- 0.05182✓
- MAE
- 0.23861✓



- MGE
- 0.02557
- MAE
- 0.01326
- MGE
- 0.01746✓
- MAE
- 0.00935✓



- MGE
- 0.02353
- MAE
- 0.00573
- MGE
- 0.00738✓
- MAE
- 0.00313✓



- MGE
- 0.03757
- MAE
- 0.01221
- MGE
- 0.01874✓
- MAE
- 0.00986✓



- MGE
- 0.01677
- MAE
- 0.00759
- MGE
- 0.01355✓
- MAE
- 0.00682✓
What the scores say
On Bench v0 (60 real photos scored against ground-truth alpha mattes at a 1024 px shorter side, lower is better), MAE is lower for withoutBG Pro Model Self-Host (0.028 vs 0.029 for withoutBG Open Weights (6% lower)); MGE (edge error) is lower for withoutBG Pro Model Self-Host (0.037 vs 0.045 for withoutBG Open Weights (18% lower)); band MAE (error near the matte edge) is lower for withoutBG Pro Model Self-Host (0.061 vs 0.067 for withoutBG Open Weights (9% lower)).
By category (MAE, categories with at least 5 images): People (12 images) withoutBG Open Weights 0.007, withoutBG Pro Model Self-Host 0.005; Objects with Details (7 images) withoutBG Open Weights 0.017, withoutBG Pro Model Self-Host 0.013; Plants (7 images) withoutBG Open Weights 0.015, withoutBG Pro Model Self-Host 0.013.
withoutBG Open Weights is open weights (Apache-2.0); withoutBG Pro Model Self-Host is proprietary (Commercial self-host license).
withoutBG models route each image to a matting model or to BiRefNet; see the router disclosure above for the split. The overall means include images with no recorded route (2 of 60 for withoutBG Open Weights, 2 of 60 for withoutBG Pro Model Self-Host), so they are not the average of the per-route means.
Specs
| withoutBG Open Weights | withoutBG Pro Model Self-Host | |
|---|---|---|
| Access | Open weights | Proprietary |
| License | Apache-2.0 | Commercial self-host license |
| Commercial use | Allowed | Per vendor terms |
| Runs on | Your own machine (ONNX, input variable) | Your own CPU servers (Docker image, no GPU needed) |
How to run each
withoutBG Open Weights
Python
pip install withoutbg
from withoutbg import WithoutBG
model = WithoutBG.open_weights()
model.remove_background("input.jpg").save("output.png")withoutBG Pro Model Self-Host
Delivered as a Docker image under a commercial license. Quickstart in the customer docs. Open the docs.
FAQ
- Which has the lower alpha error on Bench v0, withoutBG Open Weights or withoutBG Pro Model Self-Host?
- withoutBG Pro Model Self-Host has the lower MAE (withoutBG Open Weights 0.029, withoutBG Pro Model Self-Host 0.028); MGE is lower for withoutBG Pro Model Self-Host (0.037 vs 0.045 for withoutBG Open Weights (18% lower)). Scores are per-image means over 60 photos; lower is better.
- What are the license differences between withoutBG Open Weights and withoutBG Pro Model Self-Host?
- withoutBG Open Weights is open weights (Apache-2.0); withoutBG Pro Model Self-Host is proprietary (Commercial self-host license). Commercial use: withoutBG Open Weights allowed, withoutBG Pro Model Self-Host per vendor terms.
- How is the comparison scored?
- Both models run on the same 60 photos. Predictions and ground-truth mattes are scored at a 1024 px shorter side, never upscaled, with MAE (overall), MGE (Sobel gradient error, edges) and band MAE (a 10 px band around the edge). Per-image scores are published under CC BY 4.0.
Methodology
- Alpha mattes are grayscale masks encoding per-pixel foreground probability. They capture fine edge detail (hair, fur, semi-transparent regions) that binary masks discard.
- Metrics are computed by comparing predicted mattes against a ground-truth reference: MGE measures edge sharpness and MAE measures pixel-level accuracy. Both are lower-is-better.
- Scored images come from Bench v0, where every image has a ground-truth matte. The benchmark page has the full methodology and per-image scores for every model (CSV / JSON).
Photo Credits
Source images are courtesy of the photographers below via Unsplash.