- withoutBG Pro Model Self-Host
- ProprietaryLicense: Commercial self-host license
- withoutBG Pro Model API
- ProprietaryLicense: Paid cloud API
Pro Model Self-Host vs Pro Model API
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 | Pro Model Self-Host | withoutBG Pro Model API |
|---|---|---|
| MGE (Edge Quality) | 0.03692 | 0.02737 ✓ |
| MAE (Overall Accuracy) | 0.02759 | 0.02399 ✓ |
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. Pro Model API also runs an edge refiner on the result of either branch, so its BiRefNet-routed images are refined BiRefNet output, not raw BiRefNet. The table splits the scores by route: the first group is what withoutBG's own matting model handled.
| Images | Metric | Pro Model Self-Host | withoutBG Pro Model API |
|---|---|---|---|
| Routed to withoutBG matting (41) | MAE | 0.01659 | 0.01276 ✓ |
| MGE (edges) | 0.02821 | 0.01989 ✓ | |
| Routed to BiRefNet (17) | MAE | 0.01421 | 0.01233 ✓ |
| MGE (edges) | 0.03760 | 0.02601 ✓ |
Per-image averages, lower is better.



- MGE
- 0.02588
- MAE
- 0.00859
- MGE
- 0.00914✓
- MAE
- 0.00555✓



- MGE
- 0.04788
- MAE
- 0.02000
- MGE
- 0.01106✓
- MAE
- 0.00863✓
Photo by Alexander Dummer on Unsplash



- MGE
- 0.02005
- MAE
- 0.00767
- MGE
- 0.00247✓
- MAE
- 0.00143✓



- MGE
- 0.00929
- MAE
- 0.00370
- MGE
- 0.00373✓
- MAE
- 0.00172✓



- MGE
- 0.17212
- MAE
- 0.19787
- MGE
- 0.09770✓
- MAE
- 0.09947✓
Photo by Andreas Rasmussen on Unsplash



- MGE
- 0.04235✓
- MAE
- 0.00729✓
- MGE
- 0.04883
- MAE
- 0.02657



- MGE
- 0.03079
- MAE
- 0.02759✓
- MGE
- 0.02718✓
- MAE
- 0.02784



- MGE
- 0.04103
- MAE
- 0.00914
- MGE
- 0.03109✓
- MAE
- 0.00737✓



- MGE
- 0.00787✓
- MAE
- 0.00233✓
- MGE
- 0.01019
- MAE
- 0.00417



- MGE
- 0.02477
- MAE
- 0.00826
- MGE
- 0.01637✓
- MAE
- 0.00623✓



- MGE
- 0.02389
- MAE
- 0.01407
- MGE
- 0.01948✓
- MAE
- 0.01161✓



- MGE
- 0.03683
- MAE
- 0.02790✓
- MGE
- 0.02237✓
- MAE
- 0.03738



- MGE
- 0.05730
- MAE
- 0.09005
- MGE
- 0.00000✓
- MAE
- 0.00000✓
Photo by Camila Quintero Franco on Unsplash



- MGE
- 0.01250
- MAE
- 0.00669
- MGE
- 0.00847✓
- MAE
- 0.00568✓



- MGE
- 0.07267
- MAE
- 0.01349
- MGE
- 0.05622✓
- MAE
- 0.01045✓



- MGE
- 0.01582
- MAE
- 0.00245
- MGE
- 0.00297✓
- MAE
- 0.00053✓
Photo by Carlo Sierra on Unsplash



- MGE
- 0.01961
- MAE
- 0.00347
- MGE
- 0.01093✓
- MAE
- 0.00269✓



- MGE
- 0.02067
- MAE
- 0.00736
- MGE
- 0.01840✓
- MAE
- 0.00573✓



- MGE
- 0.02134✓
- MAE
- 0.00599✓
- MGE
- 0.03849
- MAE
- 0.01914
Photo by Conor Samuel on Unsplash



- MGE
- 0.02556✓
- MAE
- 0.00413✓
- MGE
- 0.02650
- MAE
- 0.00435



- MGE
- 0.01451✓
- MAE
- 0.00288✓
- MGE
- 0.03249
- MAE
- 0.00736



- MGE
- 0.01127
- MAE
- 0.00337
- MGE
- 0.00875✓
- MAE
- 0.00277✓



- MGE
- 0.02593
- MAE
- 0.00883
- MGE
- 0.01731✓
- MAE
- 0.00686✓



- MGE
- 0.05422
- MAE
- 0.01279
- MGE
- 0.03829✓
- MAE
- 0.00902✓



- MGE
- 0.01000✓
- MAE
- 0.00365✓
- MGE
- 0.01037
- MAE
- 0.00835



- MGE
- 0.01873
- MAE
- 0.00318
- MGE
- 0.00297✓
- MAE
- 0.00067✓
Photo by Lars Bo Nielsen on Unsplash



- MGE
- 0.04082
- MAE
- 0.01465✓
- MGE
- 0.01854✓
- MAE
- 0.01806
Photo by Mayer Tawfik on Unsplash



- MGE
- 0.00720✓
- MAE
- 0.00366✓
- MGE
- 0.01169
- MAE
- 0.00809



- MGE
- 0.00829✓
- MAE
- 0.00407✓
- MGE
- 0.01494
- MAE
- 0.00935



- MGE
- 0.02488
- MAE
- 0.00741
- MGE
- 0.00270✓
- MAE
- 0.00139✓
Photo by Niko Tsviliov on Unsplash



- MGE
- 0.08980
- MAE
- 0.04218
- MGE
- 0.04838✓
- MAE
- 0.04052✓
Photo by Norbert Braun on Unsplash



- MGE
- 0.00162
- MAE
- 0.00030
- MGE
- 0.00074✓
- MAE
- 0.00012✓



- MGE
- 0.05502
- MAE
- 0.02375
- MGE
- 0.04750✓
- MAE
- 0.02279✓



- MGE
- 0.02383
- MAE
- 0.02265
- MGE
- 0.01466✓
- MAE
- 0.01893✓



- MGE
- 0.02014
- MAE
- 0.01811✓
- MGE
- 0.01367✓
- MAE
- 0.01838



- MGE
- 0.02213
- MAE
- 0.00621
- MGE
- 0.01295✓
- MAE
- 0.00349✓



- MGE
- 0.36774
- MAE
- 0.49473
- MGE
- 0.33346✓
- MAE
- 0.34460✓



- MGE
- 0.02146
- MAE
- 0.00546
- MGE
- 0.01832✓
- MAE
- 0.00467✓



- MGE
- 0.00508✓
- MAE
- 0.00117✓
- MGE
- 0.00538
- MAE
- 0.00156



- MGE
- 0.00974✓
- MAE
- 0.00278✓
- MGE
- 0.01100
- MAE
- 0.00401



- MGE
- 0.04221
- MAE
- 0.00769
- MGE
- 0.03351✓
- MAE
- 0.00533✓



- MGE
- 0.02093
- MAE
- 0.02595
- MGE
- 0.01620✓
- MAE
- 0.01785✓



- MGE
- 0.01223
- MAE
- 0.00217
- MGE
- 0.01196✓
- MAE
- 0.00207✓



- MGE
- 0.00706
- MAE
- 0.00208
- MGE
- 0.00331✓
- MAE
- 0.00072✓



- MGE
- 0.01306✓
- MAE
- 0.00461✓
- MGE
- 0.01312
- MAE
- 0.00520



- MGE
- 0.01927
- MAE
- 0.00259
- MGE
- 0.00373✓
- MAE
- 0.00047✓



- MGE
- 0.02230
- MAE
- 0.00585
- MGE
- 0.01019✓
- MAE
- 0.00435✓



- MGE
- 0.01365✓
- MAE
- 0.00765✓
- MGE
- 0.02308
- MAE
- 0.01728



- MGE
- 0.13057
- MAE
- 0.02597
- MGE
- 0.07930✓
- MAE
- 0.02013✓



- MGE
- 0.07300✓
- MAE
- 0.09074✓
- MGE
- 0.07849
- MAE
- 0.09378



- MGE
- 0.12568
- MAE
- 0.02596✓
- MGE
- 0.11914✓
- MAE
- 0.02727



- MGE
- 0.02966
- MAE
- 0.00408
- MGE
- 0.00653✓
- MAE
- 0.00093✓



- MGE
- 0.00371
- MAE
- 0.00136
- MGE
- 0.00103✓
- MAE
- 0.00081✓



- MGE
- 0.00719
- MAE
- 0.01014
- MGE
- 0.00109✓
- MAE
- 0.00115✓
Photo by Vinicius Wiesehofer on Unsplash



- MGE
- 0.02527✓
- MAE
- 0.02075✓
- MGE
- 0.02893
- MAE
- 0.04267



- MGE
- 0.05182
- MAE
- 0.23861✓
- MGE
- 0.05100✓
- MAE
- 0.36202



- MGE
- 0.01746
- MAE
- 0.00935
- MGE
- 0.00127✓
- MAE
- 0.00108✓



- MGE
- 0.00738
- MAE
- 0.00313
- MGE
- 0.00481✓
- MAE
- 0.00187✓



- MGE
- 0.01874
- MAE
- 0.00986
- MGE
- 0.01195✓
- MAE
- 0.00431✓



- MGE
- 0.01355✓
- MAE
- 0.00682✓
- MGE
- 0.01769
- MAE
- 0.01273
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 API (0.024 vs 0.028 for withoutBG Pro Model Self-Host (13% lower)); MGE (edge error) is lower for withoutBG Pro Model API (0.027 vs 0.037 for withoutBG Pro Model Self-Host (26% lower)); band MAE (error near the matte edge) is lower for withoutBG Pro Model API (0.054 vs 0.061 for withoutBG Pro Model Self-Host (12% lower)).
By category (MAE, categories with at least 5 images): People (12 images) withoutBG Pro Model Self-Host 0.005, withoutBG Pro Model API 0.006; Objects with Details (7 images) withoutBG Pro Model Self-Host 0.013, withoutBG Pro Model API 0.015; Plants (7 images) withoutBG Pro Model Self-Host 0.013, withoutBG Pro Model API 0.011.
withoutBG Pro Model Self-Host is proprietary (Commercial self-host license); withoutBG Pro Model API is proprietary (Paid cloud API).
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 Pro Model Self-Host, 2 of 60 for withoutBG Pro Model API), so they are not the average of the per-route means.
Specs
| withoutBG Pro Model Self-Host | withoutBG Pro Model API | |
|---|---|---|
| Access | Proprietary | Proprietary |
| License | Commercial self-host license | Paid cloud API |
| Commercial use | Per vendor terms | Per vendor terms |
| Runs on | Your own CPU servers (Docker image, no GPU needed) | withoutBG cloud API (Frankfurt, Germany) |
How to run each
withoutBG Pro Model Self-Host
Delivered as a Docker image under a commercial license. Quickstart in the customer docs. Open the docs.
withoutBG Pro Model API
Shell
curl -X POST \
-H "X-API-Key: $WITHOUTBG_API_KEY" \
-F "file=@input.jpg" \
--output result.png \
"https://api.withoutbg.com/v1.0/image-without-background"FAQ
- Which has the lower alpha error on Bench v0, withoutBG Pro Model Self-Host or withoutBG Pro Model API?
- withoutBG Pro Model API has the lower MAE (withoutBG Pro Model Self-Host 0.028, withoutBG Pro Model API 0.024); MGE is lower for withoutBG Pro Model API (0.027 vs 0.037 for withoutBG Pro Model Self-Host (26% lower)). Scores are per-image means over 60 photos; lower is better.
- What are the license differences between withoutBG Pro Model Self-Host and withoutBG Pro Model API?
- withoutBG Pro Model Self-Host is proprietary (Commercial self-host license); withoutBG Pro Model API is proprietary (Paid cloud API). Commercial use: withoutBG Pro Model Self-Host per vendor terms, withoutBG Pro Model API 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.
When does the license pay for itself?#
The Pro Model Self-Host license is a one-off €3,600 + VAT. The Pro Model API charges 1 credit per image. Dividing the license by the per-credit price of each credit pack gives the number of images after which the API spend reaches the license price:
| API pack | Price per image | Images to reach €3,600 |
|---|---|---|
| 100 credits (€10) | €0.10 | 36,000 |
| 500 credits (€40) | €0.08 | 45,000 |
| 1,000 credits (€70) | €0.07 | 51,429 |
| 3,000 credits (€195) | €0.065 | 55,385 |
| 5,000 credits (€300) | €0.06 | 60,000 |
| 10,000 credits (€500) | €0.05 | 72,000 |
This only compares the license with API spend. It leaves out the CPU servers you run the image on, your own operations time, and VAT on the pack prices. Self-Host includes 12 months of updates; see the license terms. Current pack prices are on the pricing page.
Quality gap on Bench v0#
On the same photos, scored against ground truth (lower is better): MAE is 0.0276 for Pro Model Self-Host and 0.0240 for Pro Model API; MGE (edge error) is 0.0369 and 0.0274; band MAE is 0.0608 and 0.0536. The API runs the larger model, so the Self-Host numbers sit above it. The tables above show the same gap per image.
Deployment and privacy#
| Pro Model Self-Host | Pro Model API | |
|---|---|---|
| Runs on | Your own CPU servers (Docker image, no GPU needed) | withoutBG cloud, Frankfurt, Germany |
| Images leave your network | No | Yes, sent to the API over TLS; processed in RAM and discarded after the response |
| Billing | One-off €3,600 + VAT, perpetual, 12 months of updates | 1 credit per image, credit packs or subscription |
| Infrastructure | You operate the servers; customer-controlled infrastructure only | Managed by withoutBG |
| Smart Brush (Keep / Remove) | Yes, over its own API | Yes, over the API |