Over the past year, I’ve been the primary (and mostly only) maintainer of Albumentations.
The original team has moved on, but usage and support needs have only increased:
- Weekly issues, questions, and feature requests
- Major companies using it in production
- Sponsorships cover only 2.5% of my living expenses and 0.39% (0.0039) of what I was making as a full time employee
Companies generally don’t donate — but they do purchase licenses.
The MIT model didn’t make long-term work sustainable.
-----
AlbumentationsX uses a dual license: AGPL / Commercial
- Using AGPL or a similarly restrictive open-source license?
You can continue using AlbumentationsX under AGPL.
- For companies, three paths:
[1] Comply with AGPL by open-sourcing your full codebase
[2] Purchase a commercial license → https://albumentations.ai/pricing
[3] Keep using albumentations (MIT) — but with no updates or bug fixes
Quick comparison
albumentations (original):
- License: MIT
- Actively maintained: No
- New features: No
- Bug fixes: No
AGPL extends the GPL to network services.
If you use AGPL software in a hosted service or SaaS, you must:
- Disclose your full source code to users
- Apply AGPL to your entire codebase
This applies even if users only interact with it via an API
If your project uses MIT, Apache, or BSD — you cannot use AlbumentationsX under AGPL. You’ll need a commercial license.
-----
Why this change makes sense?
Projects like Ultralytics YOLO have successfully adopted dual licensing to:
- Fund full-time development
- Deliver new features faster
- Ensure long-term maintenance and support
My goal is the same: to work on AlbumentationsX full-time and keep it the best augmentation library available.
This model won’t suit everyone — but it enables sustainability, which the MIT model could not.
[2] Albumentations is no longer maintained.
What does this mean?
The library has been forked into a new project: AlbumentationsX. The corresponding Python package is: `albumentationsx`
It is a drop-in replacement for the original albumentations:
```bash pip uninstall albumentations pip install albumentationsx ```
Everything else stays the same:
```python import albumentations as A
transform = A.Compose([...]) ```
-----
Why?
The current situation
Over the past year, I’ve been the primary (and mostly only) maintainer of Albumentations. The original team has moved on, but usage and support needs have only increased: - Weekly issues, questions, and feature requests - Major companies using it in production - Sponsorships cover only 2.5% of my living expenses and 0.39% (0.0039) of what I was making as a full time employee
Companies generally don’t donate — but they do purchase licenses. The MIT model didn’t make long-term work sustainable.
-----
AlbumentationsX uses a dual license: AGPL / Commercial
- Using AGPL or a similarly restrictive open-source license? You can continue using AlbumentationsX under AGPL. - For companies, three paths: [1] Comply with AGPL by open-sourcing your full codebase [2] Purchase a commercial license → https://albumentations.ai/pricing [3] Keep using albumentations (MIT) — but with no updates or bug fixes
Quick comparison albumentations (original): - License: MIT - Actively maintained: No - New features: No - Bug fixes: No
albumentationsx: - License: AGPL / Commercial - Actively maintained: Yes - New features: Yes - Bug fixes: Yes - Code changes required: None
-----
What is AGPL?
AGPL extends the GPL to network services. If you use AGPL software in a hosted service or SaaS, you must:
- Disclose your full source code to users - Apply AGPL to your entire codebase
This applies even if users only interact with it via an API
If your project uses MIT, Apache, or BSD — you cannot use AlbumentationsX under AGPL. You’ll need a commercial license.
-----
Why this change makes sense?
Projects like Ultralytics YOLO have successfully adopted dual licensing to: - Fund full-time development - Deliver new features faster - Ensure long-term maintenance and support
My goal is the same: to work on AlbumentationsX full-time and keep it the best augmentation library available.
This model won’t suit everyone — but it enables sustainability, which the MIT model could not.