Black Forest Labs

BFL operates two paired entities. BFL GmbH (Amtsgericht München HRB 291820, registered at Ingeborg-Krummer-Schroth-Straße 18, 79106 Freiburg im Breisgau, with Geschäftsführer Robin Rombach, Andreas Blattmann and Axel Sauer) operates the website, impressum and German operations. Black Forest Labs Inc. (a Delaware corporation) is the named counterparty in both the consumer Terms of Service and the Developer Terms of Service -- these contracts are explicitly "governed by the laws of the State of Delaware, without regard to conflict of laws rules" with disputes resolved through binding JAMS arbitration seated in Delaware. The relationship between the two entities (parent-subsidiary, or sibling under a common holding company) is not publicly disclosed in any primary source we reviewed; both share the same management and brand.

Yes, exposure is classified as "full" for three compounding reasons. (1) Black Forest Labs Inc. is a Delaware corporation and is the named counterparty in both the consumer and developer Terms of Service; as a US-incorporated entity it is directly subject to US legal process including 18 U.S.C. § 2713 (CLOUD Act). (2) Core production infrastructure runs on Microsoft Azure -- BFL has publicly confirmed "BFL's collaboration with Microsoft started from our earliest days when we partnered with Azure to build our training and inference clusters"; the API endpoint api.bfl.ai and documentation host docs.bfl.ai are both fronted by Azure Front Door. (3) The wider stack is also US-controlled: Vercel for the web frontend (served from the Frankfurt edge), AWS CloudFront for the help center, Google Workspace for corporate email. Enterprise customers can negotiate dedicated endpoints with zero data retention and a self-hosted option that runs inference inside the customer's own environment -- both advertised on /enterprise and available on request.

BFL operates a three-tier licensing model. Apache 2.0 (fully commercial-OK): FLUX.1 [schnell] and FLUX.2 [klein] -- open weights, fork-able, redistributable. FLUX.1-dev Non-Commercial License: FLUX.1 [dev], FLUX.1 Fill / Canny / Depth / Redux [dev], FLUX.1 Kontext [dev] and FLUX.1 Krea [dev] -- open weights but research-only, no commercial use without a separate licence from BFL. Proprietary API-only: FLUX.1 [pro], FLUX.1.1 [pro] (incl. Ultra and Raw modes), FLUX.1 Kontext [pro] and the FLUX.2 [max] / [pro] / [flex] models -- these run only on the BFL API or licensed Azure AI Foundry deployments and cannot be downloaded. The GitHub repository github.com/black-forest-labs/flux ships inference code under Apache 2.0; the per-model licensing applies to the model weights, not the code.

The BFL API runs on Microsoft Azure -- both training and inference clusters are Azure-based per BFL's own disclosure. The production endpoint api.bfl.ai is fronted by Azure Front Door; the documentation host docs.bfl.ai is similarly Azure-fronted. The website itself (bfl.ai) is served by Vercel from the Frankfurt edge (fra1), and the help center sits on AWS CloudFront. The privacy policy contains only a generic EEA / SCC clause and does not name specific Azure regions; one third-party report (Sacra, not corroborated by a primary source) describes "GPU clusters in Frankfurt and Virginia for latency reduction". For deployments through Azure AI Foundry, Microsoft documents the FLUX models as available for global standard deployment in all Azure regions -- region selection in that path is delegated to the Azure customer. There is no publicly published subprocessor list, and no public Vanta trust portal we could verify; a buyer who needs binding EU-only residency should request enterprise terms.

BFL's default consumer and developer Terms reserve the right to process inputs and outputs as needed to deliver the service; the privacy policy provides a generic SCC-based transfer clause for non-EEA processing. The enterprise tier on /enterprise explicitly advertises "zero data retention and auto-scaling infrastructure" alongside dedicated endpoints and a self-hosted option ("No external calls: all inference runs inside your environment"). Customers who need a binding no-train commitment should request the enterprise tier and the available-on-request DPA. There is no publicly published subprocessor list -- "Security documentation, DPAs, and vendor assessment materials are available on request" per the enterprise page. The Responsible AI Development Policy describes content moderation, output watermarking on the hosted API, and a trust-and-safety partnership with the Internet Watch Foundation for CSAM hashing.