Watermarking of AI-generated derivative works
Herein disclosed is a method for generating and managing derivative works using generative artificial intelligence. A request to generate a derivative work based on predetermined content is received, the request comprising a requested theme. The derivative work is created as a function of the predetermined content and the requested theme using generative artificial intelligence. A watermark is embedded into the derivative work by distributing the watermark across multiple frequency bands of the derivative work, the frequency bands being less perceptible to a human ear. The watermark may comprise a unique identifier associated with the derivative work that references rights information stored in a database. Use of the derivative work may be tracked based on the unique identifier. An authorization server may be configured to govern use of the derivative work based on the watermark, including validating user requests for access based on the rights information associated with the derivative work.
1 . A method comprising:
receiving, at a generative artificial intelligence system, a request to generate a derivative audio work based on predetermined content, the request comprising a requested theme for the derivative audio work;
creating the derivative audio work as a function of the predetermined content and the requested theme by the generative artificial intelligence system; and
after creating the derivative audio work, embedding a watermark into the derivative audio work by distributing the watermark across multiple frequency bands of the derivative audio work and across a plurality of temporal segments of the derivative audio work, wherein distributing the watermark across multiple frequency bands of the derivative audio work comprises modifying frequency components of the multiple frequency bands, the multiple frequency bands being less perceptible to a human ear,
wherein the watermark comprises a payload, wherein presence of the payload in the derivative audio work is determined by the generative artificial intelligence system and indicates that the derivative audio work was generated by the generative artificial intelligence system, and wherein the payload comprises metadata that facilitates tracking of the derivative audio work and management of access to the derivative audio work.
2 . The method of claim 1 , wherein distributing the watermark across multiple frequency bands comprises a spread spectrum technique.
3 . The method of claim 1 , further comprising detecting the watermark in the derivative audio work using a detection algorithm configured to identify the watermark.
4 . The method of claim 3 , wherein the detection algorithm is configured to identify the watermark after at least one of: audio compression, format conversion, or digital editing of the derivative audio work.
5 . The method of claim 1 , wherein the watermark comprises metadata identifying at least one of: a source of the derivative audio work, a content owner identity, a date associated with the derivative audio work, usage rights associated with the derivative audio work, a model of the generative artificial intelligence system, a transformation parameter used to create the derivative audio work, an indicator that the derivative audio work is AI-generated, or provenance data documenting creation of the derivative audio work.
6 . The method of claim 1 , wherein the requested theme is converted to a text embedding in a shared latent space, and wherein creating the derivative audio work is based at least in part on the text embedding.
7 . The method of claim 1 , wherein the generative artificial intelligence system comprises a diffusion model.
8 . The method of claim 1 , wherein the predetermined content comprises audio content.
9 . A method comprising:
receiving, at a generative artificial intelligence system, a request to generate a derivative audio work based on predetermined content, the request comprising a requested theme for the derivative audio work;
creating the derivative audio work as a function of the predetermined content and the requested theme by the generative artificial intelligence system;
after creating the derivative audio work, applying a digital watermark to the derivative audio work by distributing the digital watermark across multiple frequency bands of the derivative audio work and across a plurality of temporal segments of the derivative audio work, wherein distributing the digital watermark across multiple frequency bands of the derivative audio work comprises modifying frequency components of the multiple frequency bands, the multiple frequency bands being less perceptible to a human ear, and wherein the digital watermark comprises a unique identifier associated with the derivative audio work and a payload, wherein presence of the payload in the derivative audio work is determined by the generative artificial intelligence system and indicates that the derivative audio work was generated by the generative artificial intelligence system, and wherein the payload comprises metadata that facilitates tracking of the derivative audio work and management of access to the derivative audio work;
storing rights information associated with the derivative audio work in a database, the rights information being referenced by the unique identifier; and
tracking use of the derivative audio work based on the unique identifier.
10 . The method of claim 9 , wherein the digital watermark comprises an audio watermark embedded in the derivative audio work.
11 . The method of claim 9 , wherein the digital watermark is applied using a dynamic watermarking technique, wherein the dynamic watermarking technique comprises periodically updating the digital watermark.
12 . The method of claim 9 , further comprising encrypting watermark data before applying the digital watermark to the derivative audio work.
13 . The method of claim 9 , further comprising configuring an authorization server to govern use of the derivative audio work based on the digital watermark.
14 . The method of claim 13 , wherein governing use of the derivative audio work further comprises validating user requests for access to the derivative audio work based on the rights information stored in the database.
15 . The method of claim 13 , wherein governing use of the derivative audio work further comprises revoking access to the derivative audio work upon determining that usage rights associated with the derivative audio work have expired.
16 . The method of claim 13 , wherein governing use of the derivative audio work further comprises automatically requesting an automated payment via a smart contract execution triggered based at least in part on use of the derivative audio work detected as a function of the digital watermark.
17 . The method of claim 9 , wherein the digital watermark comprises metadata identifying at least one of: a content owner identity, a date of approval, usage rights, a model of the generative artificial intelligence system, a transformation parameter used to create the derivative audio work, an indicator that the derivative audio work is AI-generated, or provenance data documenting creation of the derivative audio work.
18 . The method of claim 9 , wherein the digital watermark is imperceptible to a human observer and detectable by a detection algorithm.
19 . An article of manufacture comprising:
a non-transitory computer-readable memory storing instructions that, when executed by at least one processor, cause the at least one processor to implement a generative artificial intelligence system and perform operations comprising:
receiving, by the generative artificial intelligence system, a request to generate a derivative audio work based on predetermined content, the request comprising a requested theme for the derivative audio work;
creating the derivative audio work as a function of the predetermined content and the requested theme by the generative artificial intelligence system; and
after creating the derivative audio work, embedding a watermark into the derivative audio work by distributing the watermark across multiple frequency bands of the derivative audio work and across a plurality of temporal segments of the derivative audio work, wherein distributing the watermark across multiple frequency bands of the derivative audio work comprises modifying frequency components of the multiple frequency bands, the multiple frequency bands being less perceptible to a human ear,
wherein the watermark comprises a payload, wherein presence of the payload in the derivative audio work is determined by the generative artificial intelligence system and indicates that the derivative audio work was generated by the generative artificial intelligence system, and wherein the payload comprises metadata that facilitates tracking of the derivative audio work and management of access to the derivative audio work.