Bria's API runs in Bria's cloud by default, and that is the right choice for most teams. Enterprise customers with data-residency, sovereignty or latency requirements can also run Bria pipelines inside their own infrastructure.
Self-hosted deployment is available on Enterprise plans. Contact us to scope a deployment.
| Model | Where inference runs | Who operates it | Typical reasons |
|---|---|---|---|
| Bria Cloud | Bria's infrastructure (this API) | Bria | Fastest start, every capability, automatic updates |
| Bring Your Own Cloud (BYOC) | Your AWS, GCP or Azure account | You, with Bria packages and support | Data stays in your account, private networking, cost control at volume |
| On-premises | Your data center | You | Air-gapped or regulated environments |
Bria packages selected pipelines for self-hosting. Each ships as an installable package or container with a reference notebook, and runs without sending your images to Bria:
- Background removal (RMBG 2.0) and packshot generation
- Product dimensions images
- Object erasing (eraser) and resolution increase
- FIBO Lite image generation, including the open FIBO-VLM bridge, so the whole generation pipeline runs locally
The catalog grows over time; ask your Bria contact for the current list and hardware guidance.
Bria's model weights are also published on Hugging Face (FIBO, FIBO Edit, FIBO Lite, RMBG 2.0 and more). They are source-available for non-commercial use; commercial use requires a Bria agreement, which is what a BYOC or on-premises engagement provides.
- Provisioning. With an Enterprise plan, Bria grants your organization access to a private package registry. Access is tied to your enterprise API token, so nothing else needs to be shared.
- Install and run. Install the pipeline packages in your environment (GPU instances in your cloud account, or on-premises hardware) and run them from your own services or notebooks. Inputs and outputs never leave your environment.
- Licensing and updates. Self-hosted pipelines are licensed to your organization and use your Bria credentials to fetch updates.
- Attribution. Whenever Bria models run outside Bria's cloud, the Attribution Agent runs alongside them. It computes irreversible embeddings of generated outputs and sends only those vectors to Bria, which keeps every output attributed and covered by Bria's licensing terms without exposing your data.
You can combine models: for example, run FIBO Lite on-premises for privacy-critical generation while using Bria Cloud for editing endpoints, or generate structured prompts with the hosted /v2/structured_prompt/generate endpoint and render locally. Because VGL structured prompts are portable across the FIBO family, assets and prompts move between environments without translation.
- Attribution Agent and Service
- Safety and transparency
- Contact us to plan a deployment