BYOC — Bring Your Own Compute

Run Bria's image AI pipelines directly on your own GPU infrastructure — no calls to the hosted Bria API, no inference server to stand up. Each pipeline here is a Python package you install from Bria's private AWS CodeArtifact repository, paired with a runnable notebook that walks through setup, credentials, and a first inference end to end.

This is the right place to start if you need Bria's models running in-process, on hardware you control — for data residency, latency, air-gapped environments, or cost at scale.

Pipelines

Pipeline	What it does	Docs
Remove Background	Segments the foreground subject and returns a transparent cutout, powered by RMBG 2.0. Licensed for self-hosting.	remove-bg/
FIBO 2	Generates images from a prompt and edits images from an instruction, with structured captions (VGL) you can reuse. Licensed for self-hosting.	fibo-2/
Packshot	Produces a centered, marketplace-ready product image on a white, transparent, or custom-color background.	packshot/
Product Dimensions	Renders marketplace dimension-callout images with optional title, weight, and capacity text.	product_dimensions/
Erase	Removes a masked object and inpaints the region (LaMa coarse fill + SDXL ControlNet refine).	erase/
Increase Resolution	2×/4× super-resolution via a tiled TensorRT engine.	increase-resolution/
Fibo-Lite	Open-source image generation and structured prompt creation.	fibo-lite/
Attribution Agent	Computes image/video embeddings and reports them to the Bria API for usage attribution.	attribution-agent/

Each folder's README covers its own prerequisites, environment variables, and exact GPU/CUDA requirements — these vary per pipeline, so check the one you need before starting.

No license yet? Remove Background and FIBO 2 are sold as self-hosting licenses, per model. Each model's page is https://platform.bria.ai/licensing/pricing?model=<model>, with rmbg-2.0 or fibo-2 as the model. The other pipelines are Enterprise deliveries: talk to Bria.

How it works, in general

1. Get credentials. Every pipeline needs a BRIA_API_TOKEN: any API key of your Bria account, from the platform's API keys page. For the models Bria licenses for self-hosting (Remove Background, FIBO 2) that is all: the Bria Engine issues the package credential for the models your account holds a license for, and the model weights are downloaded from Bria on the first setup(). The other pipelines are Enterprise deliveries; their READMEs say what else they need, such as an HF_TOKEN for Hugging Face model weights.
2. Install the package from Bria's CodeArtifact repository (each pipeline names its own, e.g. bria-remove-bg, bria-packshot, bria-erase).
3. Run the notebook (code_example.ipynb in each folder) to see the pipeline load, run inference on a sample image, and produce output — then adapt it to your own images and pipeline.

Prerequisites (common to all pipelines)

• Linux with an NVIDIA GPU (exact model/VRAM requirements vary by pipeline — see each README).
• Network access to the Bria Engine and AWS CodeArtifact. The licensed models (Remove Background, FIBO 2) also download their weights from Bria's mirror on CloudFront. Hugging Face access is needed only by the pipelines whose README asks for it.
• A Bria API key (BRIA_API_TOKEN). A Hugging Face token (HF_TOKEN) only for the pipelines whose README asks for it.

export BRIA_API_TOKEN="your-api-key-here"

Getting help

If you run into issues getting a pipeline running in your environment, reach out via bria.ai/contact-us.