Remove Backgrounds with High Accuracy

Use it open-source for full privacy, or connect to our API for the sharpest, most precise cutouts.

Get API KeyCheck DocsSee Results

Want to self-host? Check our open-source version

Background removal demonstration showing before and after images of various subjects including people, products, and objects with transparent backgrounds

Remove Backgrounds with High Accuracy

Use it open-source for full privacy, or connect to our API for the sharpest, most precise cutouts.

Get API KeyCheck DocsSee Results

Want to self-host? Check our open-source version

Quick Start

Run instantly with Docker

docker run -p 80:80 withoutbg/app:latest
Press the copy button to copy this command to your clipboard

Install with Python package manager

pip install withoutbg
View on GitHub
418 stars

Please follow it or star it for the updates ⭐

Choose Your Integration Method

API, self-hosted, or local Python package

API - Model Pro

BEST QUALITY

Premium cloud API

Get API Key

Our best model, optimized for AWS Inferentia. Delivers superior pixel accuracy and edge detection with sub-second processing (p95 ≈ 800ms server-side). Premium quality for production use.

AWS Inferentia accelerated • p95 ≈ 800ms • Frankfurt, DE

API Integration Examples

POST/v1.0/image-without-background-base64
curl --request POST \
  --url https://api.withoutbg.com/v1.0/image-without-background-base64 \
  --header 'Content-Type: application/json' \
  --header 'X-API-Key: YOUR_API_KEY' \
  --data '{
  "image_base64": "/9j/4AAQSkZJRgABAQAAA..."
}'
Try DemoModel Pro ResultsAPI DocumentationPricing
DeveloperApplicationAny

Dockerized Web Application

ONE COMMAND

Self-hosted solution

View on Docker Hub

Self-hosted web interface for background removal. Run locally with one command for complete privacy. Uses Focus model by default.

Copy and run this command

docker run -p 80:80 withoutbg/app:latest
Press the copy button to copy this command to your clipboard
Focus Model ResultsDocumentationGitHub Repository
DeveloperApplicationAny (Docker)

Python SDK

OPEN SOURCE

PyPI package

View on PyPI

Install via pip for local processing or cloud API access. Focus v1.0.0 model for free local use, or connect to Pro API with api_key parameter.

Copy and run this command

pip install withoutbg
Press the copy button to copy this command to your clipboard
Focus Model ResultsGitHub RepositorySDK Documentation
DeveloperApplicationAny

Coming Soon

Native integrations

Expanding our toolkit with native integrations for popular design and editing platforms.

Blender Plugin

Coming Soon

Figma Plugin

Coming Soon

GIMP Plugin

Coming Soon

Mac App

Coming Soon

Windows App

Coming Soon

Model Pro API runs on specialized AWS Inferentia hardware for the best performance. Self-hosted options use the Focus model for complete data privacy and control.

Pricing

Remove image backgrounds with our Open Source models or Pro API at industry-leading rates (from €0.05/image). No hidden fees, no surprises.

PackageCreditsPricePrice per CreditValiditySavings
Free Trial50 credits€0€0.0030 days
Package XS100 credits€10€0.1030 days
Package S500 credits€40€0.0830 days20% off
Package M1.000 credits€70€0.0790 days30% off
Package L5.000 credits€300€0.06180 days40% off
Package XL10.000 credits€500€0.05360 days50% off

Background Removal Technology & AI Models

Comprehensive technical specifications, security protocols, and performance metrics for our AI-powered background removal API and open-source models.

AI Background Removal Models

  • Pro API Model: Premium quality background removal with advanced AI. Uses depth estimation, saliency detection, and CLIP embedding as upstream models for superior accuracy.

  • Focus Open Source Model: Most accurate open source background removal model. Uses depth estimation and saliency detection as upstream models for high-quality results.

  • Snap Open Source Model: Fastest open source background removal model. Uses depth estimation as upstream model for rapid processing.

  • All models use a refiner model as downstream task for enhanced edge quality and detail preservation.

Security & Privacy Protection

  • Trust & Transparency: Our open-source models (Focus, Snap) are Apache-2.0 licensed. Use them locally for maximum privacy control. No API calls, no data sharing.

  • API Security: TLS encryption in transit; no disk writes of customer images; no thumbnails/caches/backups stored.

  • Minimal Logging: API requests log only timestamp, endpoint, inference duration, status, request size, API key hash. No image bytes or hashes are recorded.

  • Privacy-First Analytics: No cookies on the web UI. We use Ahrefs Analytics (privacy-friendly, no cookies). To protect against abuse, we use an in-house CAPTCHA, avoiding third-party cookie-based services.

Training Data & Model Development

  • Customer Data Protection: No customer data is stored or used for training. Processing happens in RAM; buffers are zeroed after response.

  • Training Sources (Non-Customer): Public photos (e.g., Unsplash/Pexels) under permissive licenses; we created the alpha mattes. Sample dataset: withoutBG100

  • Licensed Image Sets: Rights-purchased image sets with explicit training/derivative rights for commercial use.

  • Synthetic 3D Renders: Randomized lighting/camera/focus with ground-truth mattes from the render pipeline for diverse training scenarios.

  • Studio Productions: In-studio productions targeting hard cases (hair, translucency, shadows): Creating Alpha Matting Dataset

  • Adapted Datasets: Salient-object datasets where licenses allow; re-annotated and quality-filtered for optimal training.

  • Dataset Scale: (~60K examples) image/matte pairs after QA (as of 2025-10-01). Continuously expanding dataset for improved model performance.

Technical Limitations & Considerations

  • Transparent Materials: Inconsistent results on transparent or translucent materials (e.g., glass, water, sheer fabrics) due to complex light interactions.

  • Subjective Boundaries: When foreground vs. background is subjective, results can differ. There isn't a single correct cutout. In the next release, the model will prefer objects closest to the camera (excluding ground and flat surfaces).

Anti-FAQ: The Brutal Questions

Is it really free?

Yes. Our open-source models are free under Apache-2.0. The hosted API is freemium: you get 50 free credits, then it's pay-as-you-go with credits.

  • Open source (free): Permissive Apache-2.0 license, use commercially, modify, redistribute.
  • Hosted API (freemium): Newer, more accurate models, faster inference, autoscaling, SLA.
  • Python package: Use locally with the OSS model, or call the API by adding api_key=... to the request.
  • Pricing: Buy credits; consume per image. No subscriptions or lock-in.
  • Parity: Same request/response schema, so switching between local and API is trivial.

How fast is it, end-to-end? P50/P95 latency in EU and US.

End-to-end varies with your upload/download speed, image size, and distance to our region. We therefore publish server-side processing latency (excludes network): p95 ≈ 800 ms. Servers run in Frankfurt, DE. Your total time = upload + TLS/round-trip + queueing (if any) + ~800 ms processing + download.

Do you store or train on my images or any derivatives? Exactly what (files, thumbnails, hashes, logs) is retained, for how long, where is it stored, and who can access it?

No. We don't store, index, or train on your images or any derivatives. Files are processed in RAM and discarded immediately after the response.

  • Storage: No disks, no thumbnails, no caches, no backups.
  • Processing location: Frankfurt, Germany (EU data residency).
  • Logs: We only log non-content metadata (timestamp, endpoint, duration, status code, request size, API key hash). No image bytes or hashes.
  • Security: TLS in transit; images held in memory only; buffers zeroed after processing.
  • Training: Customer data is never used to train models.

Do you use customer data?

No. We never use customer uploads for training. Customer data is never used to train models.

Supported By

NVIDIA Inception Program
AWS Activate Program