PerTh Multimodal
AI model · Safety / moderation · Text → Tables / structured data by Resemble AI
Imperceptible, multimodal watermarks for enterprise content protection and compliance.
PerTh Multimodal is an enterprise AI model that embeds imperceptible, tamper-resistant watermarks into audio, video, image, and text content. It is designed to help organizations comply with the EU AI Act while protecting intellectual property through a single, real-time API.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
Category: Cybersecurity
- Modalities
- Audio, Image, Video, Text
- Access
- REST API, SDKs
- Supported Modalities
- Audio, Image, Video, Text
- Compliance Orientation
- EU AI Act Article 50
- Processing Modes
- Real-time encode and decode
- Access Method
- Unified REST API
- Imperceptible, tamper-resistant watermarking across audio, video, image, and text
- Designed for EU AI Act Article 50 compliance requirements
- Supports real-time encoding and decoding via a single unified API
- Robust against real-world transformations like compression and re-encoding
Overview
PerTh Multimodal is an enterprise-grade watermarking model developed by Resemble AI to protect and verify AI-generated content. It extends Resemble's original PerTh technology to provide imperceptible, tamper-resistant watermarking across audio, video, image, and text modalities through a single unified API. The model is specifically designed for businesses needing to comply with the EU AI Act's requirements for marking synthetic media. Each modality utilizes a purpose-built algorithm tuned to survive real-world transformations such as compression and re-encoding. It supports both real-time encoding and decoding, making it suitable for live media streams and high-volume automated pipelines. The model can be deployed in the cloud, on-premises, or in air-gapped environments to meet various security and privacy needs.
Watermarking Capabilities
- Multimodal Watermarking: The model embeds imperceptible signatures into audio, video, image, and text content to prove authenticity and protect intellectual property.
- Tamper-Resistant Signals: Signatures are designed to survive real-world transformations including compression, re-encoding, and other common media manipulations.
- Real-Time Processing: The model supports real-time encoding and decoding of watermarks for live media streams and high-volume content pipelines.
- Purpose-Built Algorithms: Each modality uses a purpose-built algorithm with robustness tuned against specific real-world transformations for that media type.
Compliance and Integration
- EU AI Act Compliance: The model is designed to help organizations meet the requirements of Article 50 of the EU AI Act for marking AI-generated content.
- Custom Encodes: Supports explicit marks through custom encodes for audio, image, and video modalities.
- Unified API Access: The watermarking functionality is accessible through a single unified REST API for all supported modalities.
- Enterprise Security Integration: The model is part of an enterprise-grade security platform that includes deepfake detection and biometric verification.
Use cases
- Media companies embed watermarks in AI-generated video to maintain a traceable chain of custody and prove authenticity.
- Enterprises use the model to comply with EU AI Act Article 50 requirements for marking synthetic content across multiple modalities.
- Content platforms apply imperceptible signatures to audio streams to protect intellectual property against unauthorized redistribution and tampering.
- Security teams verify the provenance of digital assets by decoding watermarks that have survived compression and re-encoding.
Specifications
- Modalities
- Audio, Image, Video, Text
- Access
- REST API, SDKs
- Supported Modalities
- Audio, Image, Video, Text
- Compliance Orientation
- EU AI Act Article 50
- Processing Modes
- Real-time encode and decode
- Access Method
- Unified REST API
- Deployment Options
- Cloud, On-prem, Air-gapped
- Cloud Platforms
- AWS, Azure, Google Cloud
- Container Support
- Docker, Kubernetes
- Authentication
- SSO, SAML (Okta, Azure AD)
How it works
- API Integration: Developers integrate the watermarking capabilities into their applications using Resemble's REST API or official SDKs.
- Content Encoding: The model applies an imperceptible, tamper-resistant signature to the generated audio, video, image, or text content.
- Distribution and Transformation: The watermarked content is distributed, maintaining its traceable signature even after common transformations like compression.
- Verification and Decoding: The system uses the API to detect and decode the embedded watermark to verify the content's origin and authenticity.
Pricing
- Watermarker encode (Flex)
- $0.00050 per call
- Watermarker decode (Flex)
- $0.00020 per call
FAQ
What media types does PerTh Multimodal support?
PerTh Multimodal supports audio, video, image, and text modalities within a single unified API.
How does the watermark survive compression?
The model uses purpose-built algorithms for each modality that are tuned to be robust against common real-world transformations like compression.
Is this model compliant with the EU AI Act?
Yes, the model is specifically oriented toward helping organizations comply with Article 50 of the EU AI Act regarding the marking of AI-generated content.
How is the model integrated into applications?
Resemble AI provides a single unified REST API and official SDKs to integrate PerTh Multimodal into existing workflows.
Added by AntSpace from public sources (resemble.ai).