A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

BEN2 Base

Remove still-image backgrounds with the original BEN2 Base network at1024×1024 in an isolated browser worker.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Remove still-image backgrounds with the original BEN2 Base network at1024×1024 in an isolated browser worker.

Created by
Prama LLC
License
MIT
Model released
Unknown
Source repository created
Jan 22, 2025
Model weights · FP16
222.9 MB

Best for

  • Portrait and object foreground cutout experiments
  • Background replacement

Keep in mind

  • Independent natural portrait/dog foreground-background probes plus portrait MODNet-reference alpha consistency; not broad object, demographic or high-resolution matting certification.
  • Hair, transparency, camouflage and unfamiliar subjects can produce incorrect masks; not safety-critical segmentation.
  • This is the open Base checkpoint; the commercial enhanced model is not included.
  • Requires cross-origin isolation for eight-thread WASM and about223MB model download.
  • WebGPU official mixed precision graph failed provider assignment; no GPU support is advertised.
  • No encoded video input; upstream video helper applies the still-image model frame by frame.
  • Complete backup and source-outage recovery acceptance remain pending.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
https://huggingface.co/PramaLLC/BEN2
Release verification
Unknown
Release evidence
Original checkpoint release day not established in bounded official source review; GitHub repository was created2025-01-28. Conversion/update dates are separate.
Source updated
Dec 31, 2025
Browser model files
PramaLLC/BEN2
Files updated
Dec 31, 2025
Artifact revision
e48a20765fb421d19dcdb0bf3cc61e802ca5ec8f

Input

IMAGE

Sample result

BEN2 Base

Settings

Adds a self-contained SVG presentation image alongside the unchanged native model fields. When enabled, draws model predictions or presents generated media and text. Adds CPU postprocessing time and image bytes. It does not change model predictions.
Advanced settings 1
Keep null for a soft alpha matte, or set 0–1 to return a binary foreground mask. Lower values keep more weak predictions or foreground pixels, with more possible false positives. Higher values are stricter and may miss details. Automatic/null preserves soft alpha edges when offered.

Output

Your result will appear here

Add input or try the sample, then run the model.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

BEN2 Base

7 configurations tested
Passed7
Failed0
TestedNov 16, 2026
Runtimeonnxruntime-web 1.31.0-dev.20260914-8d85527a0

Quality criteria

{ "version": 1, "scope": "Compatibility and portrait matte fidelity on the existing annotated portrait and its mirrored form; not broad segmentation certification", "fixture": "model-samples/modnet-portrait-pexels-5965592.jpg", "reference": "test-app/public/test-images/modnet-portrait-pexels-5965592-mask.png", "minimumIoU": 0.7, "maximumAlphaMAE": 0.16, "minimumBoundaryF1": 0.6, "boundaryTolerancePixels": 12, "minimumForegroundRecall": 0.85, "maximumWarmInferenceMs": 10000, "initializationTimeoutMs": 180000, "preprocessing": "Check original BEN2.py ImageNet-normalized Lanczos1024 first; retain published onnx_run.py unnormalized bilinear alternative as a separately identified diagnostic because the two official sources conflict.", "postprocessing": "Output is already sigmoid probability. Finite min-max normalize with constant-mask guard; resize to original width,height. Do not repeat the official onnx_run.py width/height interpolation inversion.", "nativePlan": "ONNX runtime1.23.2 CPU compatibility first; original reviewed Torch2.10 CPU/MPS if needed, exact retained app-managed runtime. No Transformers5.22.1 built-in BEN2 architecture. GPU claims require measured placement.", "repetitions": "Compatibility one cold process then recommend only after repeated timing and suite; failures terminate affected case." }

PrecisionDeviceResult
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
FP16CPU · WASMPassedVery slow
Test setup & sample details
Quality gate
{ "version": 1, "scope": "Compatibility and portrait matte fidelity on the existing annotated portrait and its mirrored form; not broad segmentation certification", "fixture": "model-samples/modnet-portrait-pexels-5965592.jpg", "reference": "test-app/public/test-images/modnet-portrait-pexels-5965592-mask.png", "minimumIoU": 0.7, "maximumAlphaMAE": 0.16, "minimumBoundaryF1": 0.6, "boundaryTolerancePixels": 12, "minimumForegroundRecall": 0.85, "maximumWarmInferenceMs": 10000, "initializationTimeoutMs": 180000, "preprocessing": "Check original BEN2.py ImageNet-normalized Lanczos1024 first; retain published onnx_run.py unnormalized bilinear alternative as a separately identified diagnostic because the two official sources conflict.", "postprocessing": "Output is already sigmoid probability. Finite min-max normalize with constant-mask guard; resize to original width,height. Do not repeat the official onnx_run.py width/height interpolation inversion.", "nativePlan": "ONNX runtime1.23.2 CPU compatibility first; original reviewed Torch2.10 CPU/MPS if needed, exact retained app-managed runtime. No Transformers5.22.1 built-in BEN2 architecture. GPU claims require measured placement.", "repetitions": "Compatibility one cold process then recommend only after repeated timing and suite; failures terminate affected case." }
Tested
2026-11-16T11:44:41.571Z
Environment
Apple M3 Max /macOS27.2 /Chrome157; eight-thread cross-origin isolated WASM
Runtime
onnxruntime-web 1.31.0-dev.20260914-8d85527a0
Artifact revision
e48a20765fb421d19dcdb0bf3cc61e802ca5ec8f
Fixture
{
  "file": "model-samples/modnet-portrait-pexels-5965592.jpg",
  "size": 78124,
  "sha1": "38212739a4a4f0be25d84863cf991d7fb6c6d818",
  "sha256": "2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe",
  "url": "https://assets.localai.tools/samples/v1/files/2b5d6423b8948eb53319661633039912afa6455a9055f0c984ff88eeae1b5bfe/modnet-portrait-pexels-5965592.jpg",
  "width": 1024,
  "height": 683
}
Rerun when
model artifact revision change; Transformers.js or ONNX Runtime version change; benchmark fixture, reference matte, or quality-gate change; target browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 7050.370000001043,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 7050.255000002682,
      "inferenceMs": 7050.370000001043,
      "loadMs": 5093.434999998659,
      "totalMs": 12143.804999999702,
      "imagesPerSecond": 0.141836527728311,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait-repeat-1",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 6923.02499999851,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 6923,
      "inferenceMs": 6923.02499999851,
      "loadMs": 5093.434999998659,
      "totalMs": 12016.459999997169,
      "imagesPerSecond": 0.14444552778593392,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait-repeat-2",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 7187.72500000149,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 7187.704999998212,
      "inferenceMs": 7187.72500000149,
      "loadMs": 5093.434999998659,
      "totalMs": 12281.160000000149,
      "imagesPerSecond": 0.13912607953139453,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait-repeat-3",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 7317.780000001192,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 7317.755000002682,
      "inferenceMs": 7317.780000001192,
      "loadMs": 5093.434999998659,
      "totalMs": 12411.214999999851,
      "imagesPerSecond": 0.13665346594183442,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait-repeat-4",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 7416.030000001192,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 7416.004999998957,
      "inferenceMs": 7416.030000001192,
      "loadMs": 5093.434999998659,
      "totalMs": 12509.464999999851,
      "imagesPerSecond": 0.13484303596396444,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "id": "portrait-repeat-5",
      "iou": 0.9889790244147549,
      "boundaryF1": 0.9735125582198727,
      "foregroundRecall": 0.9920196680181895,
      "alphaMAE": 0.0045040353256968965,
      "passed": true,
      "alphaAndDimensionsValid": true,
      "inferenceMs": 7414.865000002086,
      "speedPassed": true
    },
    "runtime": {
      "durationMs": 7414.839999999851,
      "inferenceMs": 7414.865000002086,
      "loadMs": 5093.434999998659,
      "totalMs": 12508.300000000745,
      "imagesPerSecond": 0.1348642220727847,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "platform": "browser",
      "device": "wasm",
      "id": "dog",
      "reused": false,
      "probeAccuracy": 1,
      "foregroundRatio": 0.3673483535528596,
      "probes": [
        {
          "point": [
            200,
            150
          ],
          "foreground": true,
          "passed": true
        },
        {
          "point": [
            350,
            200
          ],
          "foreground": true,
          "passed": true
        },
        {
          "point": [
            500,
            400
          ],
          "foreground": true,
          "passed": true
        },
        {
          "point": [
            0,
            0
          ],
          "foreground": false,
          "passed": true
        },
        {
          "point": [
            690,
            570
          ],
          "foreground": false,
          "passed": true
        },
        {
          "point": [
            680,
            10
          ],
          "foreground": false,
          "passed": true
        }
      ],
      "passed": true,
      "runtime": {
        "durationMs": 7651.829999998212,
        "inferenceMs": 7651.954999998212,
        "loadMs": 4979.054999999702
      }
    },
    "runtime": {
      "durationMs": 7651.829999998212,
      "inferenceMs": 7651.954999998212,
      "loadMs": 4979.054999999702,
      "totalMs": 12631.009999997914,
      "imagesPerSecond": 0.1306855568283182,
      "grade": "acceptable",
      "engine": "onnxruntime-web",
      "version": "1.31.0-dev.20260914-8d85527a0",
      "actualDevice": "wasm",
      "actualPrecision": "fp16"
    }
  }
]

Something not quite right?

Tell us what happened with BEN2. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.