AI Generation Terms

The Service generates content using probabilistic artificial-intelligence models. Outputs are synthesized in response to your Inputs; they are not retrieved from a database of prior works, and they are not deterministic across runs unless you pin a seed and configuration. Different invocations of the same prompt may yield substantially different Outputs.

We make no warranty and no representation as to:

• the quality, accuracy, reliability, or fitness for purpose of any Output;
• the safety or inoffensiveness of any Output;
• the reproducibility or determinism of any Output;
• the continued availability of any particular model, style, or feature;
• the absence of similarity between any Output and pre-existing works.

You use Outputs at your own risk and are responsible for evaluating their suitability for your use.

For each Input you submit, you represent and warrant that:

• you have all rights, consents, and releases necessary to submit the Input;
• the Input does not infringe or misappropriate any third party's intellectual-property, publicity, privacy, or moral rights;
• any reference material depicting an identifiable real person is accompanied by a valid written release and proof of age, retained by you as described in the §2257 Compliance Statement;
• the Input does not violate the Content & Acceptable Use Policy.

As between you and BMG, you own the Outputs generated from your Inputs to the maximum extent permitted by applicable law, subject to the license granted to BMG in Section 5 of the Terms of Service and subject to the underlying-model licenses described in Section 10 of this document.

You acknowledge that the copyright status of AI-generated material is unsettled in many jurisdictions. In some jurisdictions (including the United States as of the effective date of these Terms), works that are entirely AI-generated may not qualify for copyright protection. We make no representation that you will obtain or can register a copyright in any Output.

Outputs are not necessarily unique. Other users may receive substantially similar Outputs from substantially similar Inputs. Ownership of an Output does not prevent that similarity.

We do not train our models or those of our vendors on Your Content without your explicit opt-in. Where you opt in, we will disclose the intended scope, purposes, and retention at the point of consent. Withdrawing consent prospectively stops use of new content for training; we will not retrieve content already incorporated into a trained model, but we will honor withdrawal for future training cycles.

Aggregated, anonymized usage metrics (such as the number of generations per hour or aggregate error rates) may be used for service improvement and capacity planning. These metrics do not contain prompt content or Output content.

The Service employs automated safety classifiers that may block, redact, modify, or flag Inputs and Outputs that appear to violate the Content Policy. Classifiers may err in either direction. False positives are an expected cost of safety operation; you may request human review by contacting [email protected]. False negatives (content that should have been blocked but was not) do not waive the Content Policy; you remain responsible for your compliance.

You may not attempt to circumvent, defeat, or test the classifiers, except in the context of an authorized security-research program.

We may add, update, modify, retrain, replace, deprecate, or remove models, styles, parameters, and features at any time, without prior notice except where these Terms or applicable law require otherwise. No particular model or feature is guaranteed to remain available. Outputs produced with a deprecated model may not be reproducible with later models.

To monitor and improve safety and reliability, a small random sample of Inputs and Outputs may be reviewed by authorized personnel or automated tooling. Where technically feasible, samples are anonymized and stripped of direct identifiers before review. You may opt out of quality-assurance sampling in your account settings; opting out does not affect moderation of current or future content, and does not reduce your obligations under the Content Policy.

You are solely responsible for how you use Outputs outside the Service, including any publication, distribution, sale, licensing, or incorporation into other works. BMG disclaims all liability arising from such downstream use. You must comply with the law of every jurisdiction in which you distribute Outputs, including any labeling, disclosure, or consent requirements applicable to AI-generated material.

The Service may incorporate or derive from open-source or commercial foundation models, each of which is subject to its own license (for example, the CreativeML Open RAIL-M license, the FLUX.1 license, or proprietary vendor licenses). The terms of those licenses apply to you in addition to these Terms to the extent they address downstream use or derivative works. We will make reasonable efforts to maintain an up-to-date list of material underlying models and their licenses, available on request at [email protected].