Responsible AI and usage guidelines for Veo

Veo on Vertex AI is designed with Google's AI principles in mind. However, it is important for developers to understand how to test and deploy Google's models safely and responsibly. To aid developers, Veo on Vertex AI has built-in safety features to help customers block potentially harmful outputs within their use cases. For more information, see safety filters.

We encourage customers to use fairness, interpretability, privacy, and security best practices when developing AI applications. For more information, see The People + AI Guidebook.

Safety filters

Veo on Vertex AI offers several ways to input prompts to generate videos, including text, video, and images. Prompts that are provided to Veo are assessed against a list of safety filters, which include harmful categories (for example, violence, sexual, derogatory, and toxic). These safety filters aim to filter out input images and videos that violate the Google Cloud Platform Acceptable Use Policy, Generative AI Prohibited Use Policy, or Our AI Principles.

If the model responds to a request with an error message, such as The prompt couldn't be submitted or it might violate our policies, then the input is triggering a safety filter. If fewer images than requested are returned, then some generated output is being blocked for not meeting safety requirements.

Safety filter code categories

Depending on the safety filters that you configure, your output may contain a safety code similar to: "Veo could not generate videos because the input image violates Vertex AI's usage guidelines. If you think this was an error, send feedback. Support codes: 15236754"

The code listed in the output corresponds to a specific harmful category.

The following table displays the error code to safety category mappings:

Error code Safety category Description
58061214
17301594 		Child Detects child content where it isn't allowed due to the API request settings or allowlisting.
29310472
15236754 		Celebrity Detects a photorealistic representation of a celebrity in the request.
62263041 Dangerous content Detects content that's potentially dangerous in nature.
57734940
22137204 		Hate Detects hate-related topics or content.
74803281
29578790
42876398 		Other Detects other miscellaneous safety issues with the request
92201652 Personal information Detects Personally Identifiable Information (PII) in the text, such as mentioning a credit card number, home addresses, or other such information.
89371032
49114662
72817394 		Prohibited content Detects the request of prohibited content in the request.
90789179
63429089
43188360 		Sexual Detects content that's sexual in nature.
78610348 Toxic Detects toxic topics or content in the text.
61493863
56562880 		Violence Detects violence-related content from the image or text.
32635315 Vulgar Detects vulgar topics or content from the text.

What's next