# US 12,182,921 B2 · Avatar style transformation using neural networks > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,182,921 B2 - Document ID: US12182921B2 - Status: Granted - Kind: B2 - Application number: 17/724,235 - Filed: 2022-04-19 - Published: 2022-08-11 - Granted: 2024-12-31 - Earliest claimed: 2018-09-19 - Assignees: Snap Inc. - Applicants: Snap Inc. - Inventors: Harrison Jesse Smith; Chen Cao; Yingying Wang - Web page: https://patentgrep.com/patent/US12182921B2 ## CPC - G06T 13/40: of characters, e.g. humans, animals or virtual beings - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - G06N 20/00: Machine learning - H04L 51/10: Multimedia information - H04L 51/04: Real-time or near real-time messaging, e.g. instant messaging [IM] - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) ## Abstract Aspects of the present disclosure involve a system comprising a computer-readable storage medium storing at least one program and a method for transforming a motion style of an avatar from a first style to a second style. The program and method include: retrieving, by a processor from a storage device, an avatar depicting motion in a first style; receiving user input selecting a second style; obtaining, based on the user input, a trained machine learning model that performs a non-linear transformation of motion from the first style to the second style; and applying the obtained trained machine learning model to the retrieved avatar to transform the avatar from depicting motion in the first style to depicting motion in the second style. ## Claims ### Claim 1 A method comprising: presenting a collection of avatars corresponding to a search criterion, each avatar in the collection of avatars being associated with metadata indicating a motion style being depicted by the avatar; determining that the motion style of a first avatar in the collection of avatars has been previously transformed; selecting a first machine learning model from two or more machine learning models based on a determination of the motion style of the first avatar; and in response to determining that the motion style of the first avatar has been previously transformed, visually distinguishing the first avatar from a second avatar in the collection of avatars. ### Claim 2 (depends on 1) The method of claim 1, further comprising: determining motion depicted by the first avatar; selecting a motion style for the first avatar; and identifying, based on the motion style, the two or more machine learning models that each perform a different non-linear transformation. ### Claim 3 (depends on 2) The method of claim 2, further comprising: blending non-linear transformation outputs of the two or more machine learning models to cause the first avatar to depict motion in the selected motion style. ### Claim 4 (depends on 1) The method of claim 1, further comprising: applying, to the first avatar, the first machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from a first motion style to a second motion style; and applying, to the first avatar, a second machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from the first motion style to a third motion style. ### Claim 5 (depends on 1) The method of claim 1, further comprising: generating a first intermediate avatar by applying the first of the two or more machine learning models to the first avatar; generating a second intermediate avatar by applying a second of the two or more machine learning models to the first avatar; and linearly interpolating the first and second intermediate avatars. ### Claim 6 (depends on 1) The method of claim 1, further comprising: storing training motions in a database, the training motions representing linear transformations of pose-by-pose motions from a first motion style to a second motion style; and training the first machine learning model based on the training motions stored in the database by generating joint positions of a single stylized pose based on joint positions of each of a subset of a plurality of the training motions. ### Claim 7 (depends on 1) The method of claim 1, further comprising: receiving a first input specifying a first amount of a first motion style; receiving a second input specifying a second amount of a second motion style; and generating a third motion style as a combination of the first amount of the first motion style and the second amount of the second motion style. ### Claim 8 A system comprising: at least one processor configured to perform operations comprising: presenting a collection of avatars corresponding to a search criterion, each avatar in the collection of avatars being associated with metadata indicating a motion style being depicted by the avatar; determining that the motion style of a first avatar in the collection of avatars has been previously transformed; selecting a first machine learning model from two or more machine learning models based on a determination of the motion style of the first avatar; and in response to determining that the motion style of the first avatar has been previously transformed, visually distinguishing the first avatar from a second avatar in the collection of avatars. ### Claim 9 (depends on 8) The system of claim 8, the operations further comprising: determining motion depicted by the first avatar; selecting a motion style for the first avatar; and identifying, based on the motion style, the two or more machine learning models that each perform a different non-linear transformation. ### Claim 10 (depends on 9) The system of claim 9, the operations further comprising: blending non-linear transformation outputs of the two or more machine learning models to cause the first avatar to depict motion in the selected motion style. ### Claim 11 (depends on 8) The system of claim 8, the operations further comprising: applying, to the first avatar, the first machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from a first motion style to a second motion style; and applying, to the first avatar, a second machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from the first motion style to a third motion style. ### Claim 12 (depends on 8) The system of claim 8, the operations further comprising: generating a first intermediate avatar by applying the first of the two or more machine learning models to the first avatar; generating a second intermediate avatar by applying a second of the two or more machine learning models to the first avatar; and linearly interpolating the first and second intermediate avatars. ### Claim 13 (depends on 8) The system of claim 8, the operations further comprising: storing training motions in a database, the training motions representing linear transformations of pose-by-pose motions from a first motion style to a second motion style; and training the first machine learning model based on the motions stored in the database by generating joint positions of a single stylized pose based on joint positions of each of a subset of a plurality of the training motions. ### Claim 14 (depends on 8) The system of claim 8, the operations further comprising: receiving a first input specifying a first amount of a first motion style; receiving a second input specifying a second amount of a second motion style; and generating a third motion style as a combination of the first amount of the first motion style and the second amount of the second motion style. ### Claim 15 A non-transitory machine-readable storage medium that includes instructions that, when executed by one or more processors of a machine, cause the machine to perform operations comprising: presenting a collection of avatars corresponding to a search criterion, each avatar in the collection of avatars being associated with metadata indicating a motion style being depicted by the avatar; determining that the motion style of a first avatar in the collection of avatars has been previously transformed; selecting a first machine learning model from two or more machine learning models based on a determination of the motion style of the first avatar; and in response to determining that the motion style of the first avatar has been previously transformed, visually distinguishing the first avatar from a second avatar in the collection of avatars. ### Claim 16 (depends on 15) The non-transitory machine-readable storage medium of claim 15, the operations further comprising: determining motion depicted by the first avatar; selecting a motion style for the first avatar; and identifying, based on the motion style, the two or more machine learning models that each perform a different non-linear transformation. ### Claim 17 (depends on 16) The non-transitory machine-readable storage medium of claim 16, the operations further comprising: blending non-linear transformation outputs of the two or more machine learning models to cause the first avatar to depict motion in the selected motion style. ### Claim 18 (depends on 15) The non-transitory machine-readable storage medium of claim 15, the operations further comprising: applying, to the first avatar, the first machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from a first motion style to a second motion style; and applying, to the first avatar, a second machine learning model of the two or more machine learning models that performs a non-linear transformation of motion from the first motion style to a third motion style. ### Claim 19 (depends on 15) The non-transitory machine-readable storage medium of claim 15, the operations further comprising: generating a first intermediate avatar by applying the first of the two or more machine learning models to the first avatar; generating a second intermediate avatar by applying a second of the two or more machine learning models to the first avatar; and linearly interpolating the first and second intermediate avatars. ### Claim 20 (depends on 15) The non-transitory machine-readable storage medium of claim 15, the operations further comprising: storing training motions in a database, the training motions representing linear transformations of pose-by-pose motions from a first motion style to a second motion style; and training the first machine learning model based on the motions stored in the database by generating joint positions of a single stylized pose based on joint positions of each of a subset of a plurality of the training motions.