- IS-Net
- Open weights · commercially permissiveLicense: Apache-2.0
- IS-Net Anime
- Open weights · commercially permissiveLicense: Apache-2.0
isnet-general-use vs isnet-anime
Quantitative comparison: 34 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 | isnet-general-use | isnet-anime |
|---|---|---|
| MGE (Edge Quality) | 0.06040 ✓ | 0.08899 |
| MAE (Overall Accuracy) | 0.04102 ✓ | 0.20329 |
- MGE
- 0.11696✓
- MAE
- 0.06654✓
- MGE
- 0.22271
- MAE
- 0.32441
- MGE
- 0.06477✓
- MAE
- 0.22343✓
- MGE
- 0.07409
- MAE
- 0.22655
- MGE
- 0.07178✓
- MAE
- 0.04442
- MGE
- 0.07609
- MAE
- 0.03315✓
- MGE
- 0.02772✓
- MAE
- 0.01521✓
- MGE
- 0.04379
- MAE
- 0.08292
- MGE
- 0.01938✓
- MAE
- 0.00368✓
- MGE
- 0.06757
- MAE
- 0.03756
- MGE
- 0.06178✓
- MAE
- 0.02413✓
- MGE
- 0.07908
- MAE
- 0.35433
- MGE
- 0.03584✓
- MAE
- 0.02595✓
- MGE
- 0.04128
- MAE
- 0.05713
- MGE
- 0.07545✓
- MAE
- 0.01532✓
- MGE
- 0.15451
- MAE
- 0.05766
- MGE
- 0.03334✓
- MAE
- 0.05553✓
- MGE
- 0.09780
- MAE
- 0.36902
- MGE
- 0.05666✓
- MAE
- 0.02750✓
- MGE
- 0.06330
- MAE
- 0.09736
- MGE
- 0.05658✓
- MAE
- 0.01533✓
- MGE
- 0.09962
- MAE
- 0.09069
- MGE
- 0.04454✓
- MAE
- 0.01054✓
- MGE
- 0.11255
- MAE
- 0.26121
- MGE
- 0.01855✓
- MAE
- 0.00670✓
- MGE
- 0.02415
- MAE
- 0.01178
- MGE
- 0.03953✓
- MAE
- 0.01182✓
- MGE
- 0.08268
- MAE
- 0.03511
- MGE
- 0.10810✓
- MAE
- 0.08165✓
- MGE
- 0.16656
- MAE
- 0.20372
- MGE
- 0.05057
- MAE
- 0.02811✓
- MGE
- 0.04387✓
- MAE
- 0.04381
- MGE
- 0.03985
- MAE
- 0.02953✓
- MGE
- 0.03303✓
- MAE
- 0.48336
- MGE
- 0.19151✓
- MAE
- 0.25475✓
- MGE
- 0.22455
- MAE
- 0.42526
- MGE
- 0.07476✓
- MAE
- 0.04964✓
- MGE
- 0.07686
- MAE
- 0.06173
- MGE
- 0.03981✓
- MAE
- 0.01439✓
- MGE
- 0.13809
- MAE
- 0.27817
- MGE
- 0.01708✓
- MAE
- 0.01406✓
- MGE
- 0.02504
- MAE
- 0.02957
- MGE
- 0.03986
- MAE
- 0.02022✓
- MGE
- 0.03548✓
- MAE
- 0.02746
- MGE
- 0.04084✓
- MAE
- 0.00708✓
- MGE
- 0.12355
- MAE
- 0.22925
- MGE
- 0.03088✓
- MAE
- 0.03504✓
- MGE
- 0.06489
- MAE
- 0.18421
- MGE
- 0.02121
- MAE
- 0.00439✓
- MGE
- 0.01706✓
- MAE
- 0.00473
- MGE
- 0.01959✓
- MAE
- 0.00611✓
- MGE
- 0.03663
- MAE
- 0.32221
- MGE
- 0.04488✓
- MAE
- 0.02536✓
- MGE
- 0.05477
- MAE
- 0.17787
- MGE
- 0.07203
- MAE
- 0.02753✓
- MGE
- 0.05636✓
- MAE
- 0.56529
- MGE
- 0.11521
- MAE
- 0.06318✓
- MGE
- 0.08707✓
- MAE
- 0.58444
- MGE
- 0.11301✓
- MAE
- 0.02857✓
- MGE
- 0.13132
- MAE
- 0.05145
- MGE
- 0.09085✓
- MAE
- 0.01698✓
- MGE
- 0.26981
- MAE
- 0.19284
- MGE
- 0.07211
- MAE
- 0.02521✓
- MGE
- 0.07205✓
- MAE
- 0.42163
- MGE
- 0.11379
- MAE
- 0.08696✓
- MGE
- 0.08625✓
- MAE
- 0.50563
- MGE
- 0.03462✓
- MAE
- 0.02967✓
- MGE
- 0.04334
- MAE
- 0.08020
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).