AWS ai medium security documentation change
Summary
Updated release date, added Reel 1.1 capabilities with multi-shot modes, enhanced content moderation details, compliance references to EU AI Act and Frontier Safety Commitments, revised security controls documentation, and updated performance metrics for harmful content filtering
Security assessment
Added explicit compliance requirements with EU AI Act and Frontier Safety Commitments, documented improved content moderation metrics (96.4% harmful prompt blocking), and expanded security controls for encryption/IAM. These changes address security vulnerabilities related to harmful content generation and regulatory compliance.
Diff
diff --git a/ai/responsible-ai/nova-reel/overview.md b/ai/responsible-ai/nova-reel/overview.md index 241ff9891..ee859c7de 100644 --- a//ai/responsible-ai/nova-reel/overview.md +++ b//ai/responsible-ai/nova-reel/overview.md @@ -15 +15 @@ An AWS AI Service Card explains the use cases for which the service is intended, -This Service Card applies to the release of Amazon Nova Reel that is current as of December 3, 2024. +This Service Card applies to the release of Amazon Nova Reel that is current as of April 7, 2025. @@ -19 +19 @@ This Service Card applies to the release of Amazon Nova Reel that is current as -Amazon Nova Reel is a proprietary multimodal foundation model (FM) designed for enterprise use cases. Amazon Nova Reel generates a novel video from a descriptive natural language text string and an optional reference image (together, the “prompt”). Customers can use Amazon Nova Reel to create content within advertising, branding, product design, book illustration, home design, fashion mock-up, and social media workflows. This AI Service Card applies to the use of Amazon Nova Reel via [ Amazon Bedrock Console](https://docs.aws.amazon.com/bedrock/latest/userguide/using-console.html) and [ Amazon Bedrock API](https://awscli.amazonaws.com/v2/documentation/api/latest/reference/bedrock/index.html). Typically, customers use the Console to develop and test applications, and the API for production loads at scale. Each Nova model is a managed subservice of Amazon Bedrock; customers can focus on executing prompts without having to provision or manage any infrastructure such as instance types, network topology, and endpoints. +Amazon Nova Reel is a proprietary multimodal foundation model (FM) designed for enterprise use cases. Amazon Nova Reel generates a novel video from a descriptive natural language text string and an optional reference image (together, the “prompt”). Customers can use Amazon Nova Reel to create content within advertising, branding, product design, and social media workflows. This AI Service Card applies to the use of Amazon Nova Reel via [ Amazon Bedrock Console](https://docs.aws.amazon.com/bedrock/latest/userguide/using-console.html) and [ Amazon Bedrock API](https://awscli.amazonaws.com/v2/documentation/api/latest/reference/bedrock/index.html). Typically, customers use the Console to develop and test applications, and the API for production loads at scale. Each Nova model is a managed subservice of Amazon Bedrock; customers can focus on executing prompts without having to provision or manage any infrastructure such as instance types, network topology, and endpoints. @@ -30,0 +31,2 @@ Amazon Nova Reel serves a wide range of potential application domains and offers + * Reel 1.0 + @@ -34,0 +37,6 @@ Amazon Nova Reel serves a wide range of potential application domains and offers + * Reel 1.1 + + * Generate videos of up to 120 seconds given a text prompt. + + * Generate videos of up to 120 seconds given a reference image and a text prompt. + @@ -42 +50,9 @@ When assessing a video generation model for a particular use case, we encourage -For example, Amazon Nova Reel can be used as a creative tool to help advertisers or brands create video-based assets for their advertising or marketing campaigns. The business goal may be to improve the cost, quality, and productivity required to create a video asset to be used in marketing campaigns. The stakeholders may include the advertiser or brand, who wants to create a functional video Ad. The workflow is 1/ the advertiser or brand provides a text prompt that will be suited for the video asset or a product image that needs to be converted into a small video, and 2/ based on the text prompt and/or input product images, the customer can iterate with Amazon Nova Reel to create video assets based on their relevant marketing message using text-to-video and image-to-video capabilities. Output videos may contain the product image provided by the user, depictions of the details mentioned in the prompt, as well as other components that the model fills in. Input variations include all the normal variations in English expression across different individuals, differences in the definition of design concepts/jargon, inaccuracies, misspellings, and undefined abbreviations. The error types, ranked in order of estimated negative impact on stakeholders, include a/ harmful or otherwise inappropriate content, and b/ misrepresenting the input product. +For example, Amazon Nova Reel can be used as a creative tool to help advertisers or brands create video-based assets for their advertising or marketing campaigns. The business goal may be to improve the cost, quality, and productivity required to create a video asset to be used in marketing campaigns. The stakeholders may include the advertiser or brand, who wants to create a functional video ad. + +Amazon Nova Reel 1.0 offers only Single Shot mode. Amazon Nova Reel 1.1 offers three modes: 1/ Single Shot, 2/ Multi-shot Automated, and 3/ Multi-shot Manual + +In Single Shot mode, customers can create a 6-second video asset using only a text prompt or a text prompt in combination with an input image. The workflow is the customer 1/ provides a text prompt describing the desired video, and 2/ provides an image to be used as the starting frame for the video (optional). + +In Multi-shot Automated mode, customers have the ability to create a video asset using a single text prompt. The workflow is the customer 1/ provides a text prompt describing the desired video, and 2/ specifies the desired output video duration. The output will be a long video comprising multiple shots that reflect the text prompt and video duration delivered as 1/ a single stitched video in the specified duration, and 2/ individual 6 second shots stored separately in the customer’s storage location. Shots are fixed at 6 seconds. If the input or resulting video violates [our core dimensions of Responsible AI,](https://aws.amazon.com/ai/responsible-ai/) the model will not generate the video + +In Multi-shot Manual mode, customers have the opportunity to provide a text prompt and optional input image for each shot within the long video, allowing granular controllability which is essential for precise storytelling and creative direction. The workflow is the customer 1/ provides a series of text prompts, one for each shot, and 2/ provides images to be used as the starting frames for one or more of those shots (optional). The output will be a long video comprising multiple shots that reflect the text prompts and images provided for each shot, delivered as 1/ single stitched video, and 2/ individual 6 second shots stored separately in the customer’s storage location. Shots are fixed at 6 seconds. If the input or resulting video violates [our core dimensions of Responsible AI,](https://aws.amazon.com/ai/responsible-ai/) the model will not generate the video Output videos will contain any images provided as input by the user, depictions of the details mentioned in the prompt, as well as other components that the model fills in. Input variations include all the normal variations in English expression across different individuals, differences in the definition of design concepts/jargon, inaccuracies, misspellings, and undefined abbreviations. The error types, ranked in order of estimated negative impact on stakeholders, include a/ harmful or otherwise inappropriate content, and b/ misrepresenting the input product. @@ -47 +63 @@ For example, Amazon Nova Reel can be used as a creative tool to help advertisers -_"Cinematic dolly shot of a juicy cheeseburger with melting cheese, fries, and a condensation-covered cola on a worn diner table. Natural lighting, visible steam and droplets. 4k, photorealistic, shallow depth of field"_ +_"A cute raccoon playing guitar underwater."_ @@ -57 +73 @@ Amazon Nova Reel has a number of limitations requiring careful consideration. -We make every effort to design, develop, and rigorously test our models to help ensure they produce appropriate outputs based on user inputs, but generative models are by their nature non-deterministic and may occasionally produce unintended or undesirable outputs. We encourage users to report questions and provide feedback here about our models to help us continuously improve their performance. Customers should evaluate outputs for accuracy and appropriateness for their use case, especially if they will be directly surfaced to end users. Additionally, if Amazon Nova Reel is used in customer workflows that produce consequential decisions, customers must evaluate the potential risks of their use case and implement appropriate human oversight, testing, and other use case-specific safeguards to mitigate such risks. See the [AWS Responsible AI Policy](https://aws.amazon.com/ai/responsible-ai/policy/) for more information. +We make every effort to design, develop, and rigorously test our models to help ensure they produce appropriate outputs based on user inputs, but generative models are by their nature non-deterministic and may occasionally produce unintended or undesirable outputs. We encourage users to report questions and provide feedback [here](https://pages.awscloud.com/global-ln-gc-400-ai-service-cards-contact-us-registration.html) about our models to help us continuously improve their performance. Customers should evaluate outputs for accuracy and appropriateness for their use case, especially if they will be directly surfaced to end users. Additionally, if Amazon Nova Reel is used in customer workflows that produce consequential decisions, customers must evaluate the potential risks of their use case and implement appropriate human oversight, testing, and other use case-specific safeguards to mitigate such risks. See the [AWS Responsible AI Policy](https://aws.amazon.com/ai/responsible-ai/policy/) for more information. Customers who use Amazon Nova Reel models are responsible for ensuring that their use of Amazon Nova Reel and the generated video or other output complies with all applicable laws. The model and output may not be used for any prohibited practices under the EU AI Act. @@ -67 +83 @@ Amazon Nova Reel is designed to disengage with attempts to circumvent its safety -Amazon Nova Reel text prompts cannot exceed 512 characters. Additionally, input images are limited to specific dimensions (only 1280x720). Reference images, which are submitted as part of the prompt, can be formatted as either PNG or JPEG. For more information, see [Amazon Nova User Guide](https://docs.aws.amazon.com/nova/latest/userguide). +Amazon Nova Reel text prompts for Single Shot or Multi-shot Manual modes cannot exceed 512 characters. Prompts for Multi-shot Automated model cannot exceed 4,000 characters. Additionally, input images are limited to specific dimensions (only 1280x720). Reference images, which are submitted as part of the prompt, can be formatted as either PNG or JPEG. For more information, see [Amazon Nova User Guide](https://docs.aws.amazon.com/nova/latest/userguide). @@ -82 +98 @@ Amazon Nova Reel is trained from images and videos of objects. It does not store -Customers can influence the composition of a generated video via detailed prompting. However, customers should not expect to be able to describe all aspects of any desired video with a 512-character text prompt, e.g., there are many possible generated videos that would match a text prompt of '_the dog._ ' Amazon Nova Reel "fills in" unspecified information automatically, extrapolating creatively from images and videos. Thus, customers may encounter unexpected elements in generated outputs, such as unrealistic shapes with impossible angles or proportions, inconsistent lighting, or unnatural colorations. +Customers can influence the composition of a generated video via detailed prompting. However, customers should not expect to be able to describe all aspects of any desired video within the maximum character length of a text prompt, e.g., there are many possible generated videos that would match a text prompt of '_the dog._ ' Amazon Nova Reel "fills in" unspecified information automatically, extrapolating creatively from images and videos. Thus, customers may encounter unexpected elements in generated outputs, such as unrealistic shapes with impossible angles or proportions, inconsistent lighting, or unnatural colorations. @@ -92 +108 @@ There are a wide variety of video styles (for example, illustration, digital, fi -Amazon Nova Reel is based on an ML technology (diffusion model) that does not explicitly model parts of objects. As a result, it might produce depictions of the human face and body that are anatomically incorrect (for example, noses, fingers and toes). Customers who use Amazon Nova Canvas to generate humans are responsible for ensuring that the output and use of the generated video complies with all applicable laws, including but not limited to laws governing biometric privacy or digital replicas. +Amazon Nova Reel is based on an ML technology (diffusion model) that does not explicitly model parts of objects. As a result, it might produce depictions of the human face and body that are anatomically incorrect (for example, noses, fingers and toes). Customers who use Amazon Nova Reel to generate humans are responsible for ensuring that the output and use of the generated video complies with all applicable laws, including but not limited to laws governing biometric privacy or digital replicas. @@ -94 +110 @@ Amazon Nova Reel is based on an ML technology (diffusion model) that does not ex -**Scene Text** +**On-screen Text** @@ -104 +120 @@ Amazon Nova Reel currently only supports video output generation for English pro -**Image Output** +**Output Resolution and Framerate** @@ -107 +123 @@ Amazon Nova Reel currently only supports video output generation for English pro -For a list of supported resolutions, see [Amazon Nova User Guide](https://docs.aws.amazon.com/nova/latest/userguide). +For a list of supported resolutions and framerates, see [Amazon Nova User Guide](https://docs.aws.amazon.com/nova/latest/userguide). @@ -138 +154 @@ We use multiple datasets and human work forces to evaluate the performance of No - * _Independent Red Teaming Network:_ In accordance with our commitment to the US White House on ensuring Safe, Secure, and Trustworthy AI, we partner with a variety of third parties to conduct red teaming against our AI models. We leverage red teaming firms to complement our in-house testing in areas such as safety, security, privacy, fairness, and veracity-related topics. We also work with specialized firms and academics to red-team our models for specialized areas such as Cybersecurity and Chemical, Biological, Radiological, and Nuclear (CBRN) capabilities. + * _Independent Red Teaming Network:_ Independent Red Teaming Network: Consistent with our Frontier AI Safety Commitments on ensuring Safe, Secure, and Trustworthy AI, we partner with a variety of third parties to conduct red teaming against our AI models. We leverage red teaming firms to complement our in-house testing in areas such as safety, security, privacy, fairness, and veracity-related topics. We also work with specialized firms and academics to red-team our models for specialized areas such as Cybersecurity and Chemical, Biological, Radiological, and Nuclear (CBRN) capabilities. @@ -148 +164 @@ Safety is a shared responsibility between AWS and our customers. Our goal for sa - * _Harmlessness:_ We evaluate Amazon Nova Reel's ability to accurately reject potentially harmful prompts using multiple datasets. For example, on a proprietary dataset (1k samples) containing prompts that attempt to solicit videos containing harmful content (e.g., abuse, violence, hate, nudity, insults, profanity), Amazon Nova Reel correctly blocks 99.4% of harmful prompts. In order to ensure we are maintaining high performance, we augment our training dataset with benign prompts, and we measure our true pass rate for harmless prompts using 1/ MS-COCO Uni-grams, Bi-grams, and Tri-grams test set with a 98.2% pass rate, and 2/ an internally curated set of common-nouns and short phrases with a 98.2% pass rate. + * _Harmlessness:_ We evaluate Amazon Nova Reel 1.1's ability to accurately reject potentially harmful prompts using multiple datasets. For example, on a proprietary dataset (1.3k samples) containing prompts that attempt to solicit videos containing harmful content (e.g., abuse, violence, hate, nudity, insults, profanity), Amazon Nova Reel 1.1 correctly blocks 96.4% of harmful prompts. In order to ensure we are maintaining high performance, we augment our training dataset with benign prompts, and we measure our true pass rate for harmless prompts using an internally curated set of common-nouns andphrases with a 92.1% pass rate. @@ -150 +166 @@ Safety is a shared responsibility between AWS and our customers. Our goal for sa - * _Toxicity:_ Toxicity is a common but narrow form of harmfulness, on which individual opinion varies widely. We assess our ability to avoid prompts and to not generate videos that contain potentially toxic content using automated testing with multiple datasets, and find that Amazon Nova Reel performs well on common types of toxicity. For example, on a proprietary toxic image-prompt dataset (0.6k samples), which we classified into sub-categories (e.g., violence, gore, self-harm), Amazon Nova Reel's end-to-end toxicity guardrails accurately block 99.4% of toxic content. + * _Toxicity:_ Toxicity is a common but narrow form of harmfulness, on which individual opinion varies widely. We assess our ability to avoid prompts and to not generate videos that contain potentially toxic content using automated testing with multiple datasets, and find that Amazon Nova Reel 1.1 performs well on common types of toxicity. For example, on a proprietary toxic image-prompt dataset (3.4k samples), which we classified into sub-categories (e.g., violence, gore, self-harm), Amazon Nova Reel 1.1’s end-to-end toxicity guardrails accurately block 95.8% of toxic content. @@ -152 +168 @@ Safety is a shared responsibility between AWS and our customers. Our goal for sa - * _Chemical, Biological, Radiological, and Nuclear (CBRN):_ Compared to information available via internet searches, science articles, and paid experts, we see no indications that Amazon Nova Reel increases access to information about chemical, biological, radiological or nuclear threats. However, we will continue testing, and per the voluntary [ AI White House commitments](https://www.aboutamazon.com/news/company-news/amazon-responsible-ai), will engage with other video generator vendors to share, learn about, and mitigate possible CBRN threats and vulnerabilities. + * _Chemical, Biological, Radiological, and Nuclear (CBRN)_ : Compared to information available via internet searches, science articles, and paid experts, we see no indications that Amazon Nova Reel increases access to information about chemical, biological, radiological or nuclear threats. Consistent with our voluntary endorsement of the Frontier AI Safety Commitments at the AI Seoul Summit, we continue to test for CBRN risk, and engage with other video generator vendors to share, learn about, and mitigate possible CBRN threats and vulnerabilities. @@ -168 +184 @@ Amazon Nova Reel is designed to generate videos that a diverse set of customers - 1. It is not currently possible to build training datasets that cover all varieties of every object; however, for humans in particular, we aim to combat societal bias and cultural appropriation. We test Amazon Nova Reel ability to moderate these outcomes using a proprietary dataset of aggregated red teaming iterations that depict bias, stereotyping, and hate against individuals and groups. We find that Amazon Nova Reel blocks 97.4% of observed bias in generations. + 1. It is not currently possible to build training datasets that cover all varieties of every object; however, for humans in particular, we aim to combat societal bias and cultural appropriation. We test Amazon Nova Reel ability to moderate these outcomes using a proprietary dataset of aggregated red teaming iterations that depict bias, stereotyping, and hate against individuals and groups. We find that Amazon Nova Reel 1.1 blocks 95.2% of observed bias in generations. @@ -170 +186 @@ Amazon Nova Reel is designed to generate videos that a diverse set of customers - 2. When users provide no guidance about the desired attributes of an object or person, it is unclear how to judge output over repeated renderings of the object. For example, for the prompt "basketball players", some users might prefer a team with similar demographic attributes and other might want a distribution of attributes (for example, gender) matching some distribution they have in mind. Given this ambiguity, when there is no information included in the prompt, Amazon Nova Reel is designed to return diverse results, but without specifying a desired distribution. For example, on a proprietary dataset used to test rendering of retail products, we find that when no gender nouns or pronouns are present in a text prompt requiring a human face, the model generates female faces 52% of the time and male faces 48% of the time. For a different prompt dataset asking for images of people in 14 occupations (for example, CEO, teacher, judge, social worker, cashier), we find that gender disparity is less than 5% for all 14 occupations. Given the current limits of datasets and technology, and the intrinsic ambiguity of generating videos without guidance, we recommend that customers consider specifying desired object attributes in the text prompt. + 2. When users provide no guidance about the desired attributes of an object or person, it is unclear how to judge output over repeated renderings of the object. For example, for the prompt "basketball players", some users might prefer a team with similar demographic attributes and other might want a distribution of attributes (for example, gender) matching some distribution they have in mind. Given this ambiguity, when there is no information included in the prompt, Amazon Nova Reel is designed to return diverse results, but without specifying a desired distribution. Given the current limits of datasets and technology, and the intrinsic ambiguity of generating videos without guidance, we recommend that customers consider specifying desired object attributes in the text prompt. @@ -203 +219 @@ Amazon Nova Reel is available in Amazon Bedrock. Amazon Bedrock is a managed ser -All Amazon Bedrock models, including Amazon Nova Reel, come with enterprise security that enables customers to build generative AI applications that support common data security and compliance standards, including GDPR and HIPAA. Customers can use AWS PrivateLink to establish private connectivity between customized Titan models and on-premises networks without exposing customer traffic to the internet. Customer data is always encrypted in transit and at rest, and customers can use their own keys to encrypt the data, for example, using AWS Key Management Service (AWS KMS). Customers can use AWS Identity and Access Management (IAM) to securely control access to Amazon Bedrock resources. Also, Amazon Bedrock offers comprehensive monitoring and logging capabilities that can support customer governance and audit requirements. For example, Amazon CloudWatch; can help track usage metrics that are required for audit purposes, and AWS CloudTrail can help monitor API activity and troubleshoot issues as Amazon Nova Reel is integrated with other AWS systems. Customers can also choose to store the metadata, prompts, and video generations in their own encrypted Amazon Simple Storage Service (Amazon S3) bucket. For more information, see [BRlong; Security](https://docs.aws.amazon.com/bedrock/latest/userguide/security.html). +All Amazon Bedrock models, including Amazon Nova Reel, come with enterprise security that enables customers to build generative AI applications that support common data security and compliance standards, including GDPR and HIPAA. Customers can use AWS PrivateLink to establish private connectivity between customized Titan models and on-premises networks without exposing customer traffic to the internet. Customer data is always encrypted in transit and at rest, and customers can use their own keys to encrypt the data, for example, using AWS Key Management Service (AWS KMS). Customers can use AWS Identity and Access Management (IAM) to securely control access to Amazon Bedrock resources. Also, Amazon Bedrock offers comprehensive monitoring and logging capabilities that can support customer governance and audit requirements. For example, Amazon CloudWatch; can help track usage metrics that are required for audit purposes, and AWS CloudTrail can help monitor API activity and troubleshoot issues as Amazon Nova Reel is integrated with other AWS systems. Customers can also choose to store the metadata, prompts, and video generations in their own encrypted Amazon Simple Storage Service (Amazon S3) bucket. For more information, see [Amazon Bedrock Security](https://docs.aws.amazon.com/bedrock/latest/userguide/security.html). @@ -244 +260 @@ The performance of any application using Amazon Nova Reel depends on the design - 3. _Scene text:_ When trying to display text elements in the image generation, Amazon Nova Reel produces better results when provided with double quotes in the prompt. For example, _'an image of a boy holding a sign that says "success"'_ instead of _'an image of a boy holding a sign that says success'_. + 3. _On-screen text:_ When trying to display text elements in the image generation, Amazon Nova Reel produces better results when provided with double quotes in the prompt. For example, _'an image of a boy holding a sign that says "success"'_ instead of _'an image of a boy holding a sign that says success'_.