# US 2026/0094370 A1 · Creating three-dimensional (3d) avatars

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0094370 A1
- Document ID: US20260094370A1
- Status: Published application
- Kind: A1
- Application number: 18/902,805
- Filed: 2024-09-30
- Published: 2026-04-02
- Earliest claimed: 2024-09-30
- Assignees: Beijing Zitiao Network Technology Co., Ltd.; Lemon Inc.
- Applicants: Beijing Zitiao Network Technology Co., Ltd.; Lemon Inc.
- Inventors: Jianfeng Zhang; Huichao Zhang; Jun Hao Liew; Chenxu Zhang; Jiashi Feng
- Web page: https://patentgrep.com/patent/US20260094370A1

## CPC

- G06T 17/20: Finite element generation, e.g. wire-frame surface description, {tesselation}
- G06T 13/40: of characters, e.g. humans, animals or virtual beings
- G06T 15/20: Perspective computation
- G06T 17/00: Three-dimensional [3D] modelling for computer graphics
- G06T 19/20: Editing of three-dimensional [3D] images, e.g. changing shapes or colours, aligning objects or positioning parts
- G06T 2219/2021: Shape modification
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

Techniques for creating high-quality and animatable three-dimensional (3D) avatars are provided. A 3D human representation with a pre-defined pose is generated in a canonical space by adopting a coarse-to-fine pipeline. The 3D human representation corresponds to a 3D avatar to be created based on input text. The 3D human representation is deformed to a target pose defined by a Skinned Multi-Person Linear (SMPL) parameter in a deformed space. The 3D avatar is created by implementing iterative optimization of the 3D human representation in the canonical space and the deformed space by applying a Score Distillation Sampling (SDS) loss based on the input text and information indicative of poses.

## Claims

### Claim 1

A method of creating three-dimensional (3D) avatars, comprising:
generating a 3D human representation with a pre-defined pose in a canonical space by adopting a coarse-to-fine pipeline, wherein the 3D human representation corresponds to a 3D avatar to be created based on input text;
deforming the 3D human representation to a target pose defined by a Skinned Multi-Person Linear (SMPL) parameter in a deformed space; and
creating the 3D avatar by implementing iterative optimization of the 3D human representation in the canonical space and the deformed space by applying a Score Distillation Sampling (SDS) loss based on the input text and information indicative of poses, wherein the poses comprise the pre-defined pose and the target pose.

### Claim 2 (depends on 1)

The method of claim 1, wherein the coarse-to-fine pipeline comprises a coarse stage and a fine stage, and wherein the generating a 3D human representation comprises:
learning a static human representation with the pre-defined pose by leveraging low-resolution diffusion prior using a first sub-model in the coarse stage;
initializing a second sub-model with a mesh resulted from the coarse stage; and
modeling the 3D human representation as a textured mesh with high-resolution diffusion prior by using the second sub-model in the fine stage.

### Claim 3 (depends on 1)

The method of claim 1, further comprising:
refining texture and geometric details of the 3D human representation based on zooming in on body parts of the 3D human representation and applying the SDS loss.

### Claim 4 (depends on 1)

The method of claim 1, further comprising:
rendering a plurality of images of the 3D human representation from a plurality of camera viewpoints.

### Claim 5 (depends on 4)

The method of claim 4, wherein the plurality of images comprising a full-body image, a head-part image, a leg-part image, and an arm-part image.

### Claim 6 (depends on 4)

The method of claim 4, further comprising:
rendering a plurality of meshes corresponding to the plurality of images from the plurality of camera viewpoints; and
utilizing the plurality of meshes as conditions for the iterative optimization to ensure view-consistency and to enable accurate pose control of the created 3D avatar.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
enhancing pose control generalization and accuracy by applying different poses in the deformed space.

### Claim 8 (depends on 1)

The method of claim 1, further comprising:
receiving a reference image indicative of a particular style; and
creating the 3D avatar by applying the SDS loss based on the input text, the information indicative of the poses, and the reference image, wherein the created 3D avatar has the particular style.

### Claim 9 (depends on 1)

The method of claim 1, further comprising:
animating the 3D avatar based on estimating SMPL sequences from a driving video; or
animating the 3D avatar based on converting text into SMPL sequences, wherein the text describes one or more motions.

### Claim 10

A system of creating three-dimensional (3D) avatars, comprising:
at least one processor; and
at least one memory communicatively coupled to the at least one processor and comprising computer-readable instructions that upon execution by the at least one processor cause the at least one processor to perform operations comprising:
generating a 3D human representation with a pre-defined pose in a canonical space by adopting a coarse-to-fine pipeline, wherein the 3D human representation corresponds to a 3D avatar to be created based on input text;
deforming the 3D human representation to a target pose defined by a Skinned Multi-Person Linear (SMPL) parameter in a deformed space; and
creating the 3D avatar by implementing iterative optimization of the 3D human representation in the canonical space and the deformed space by applying a Score Distillation Sampling (SDS) loss based on the input text and information indicative of poses, wherein the poses comprise the pre-defined pose and the target pose.

### Claim 11 (depends on 10)

The system of claim 10, wherein the coarse-to-fine pipeline comprises a coarse stage and a fine stage, and wherein the generating a 3D human representation comprises:
learning a static human representation with the pre-defined pose by leveraging low-resolution diffusion prior using a first sub-model in the coarse stage;
initializing a second sub-model with a mesh resulted from the coarse stage; and
modeling the 3D human representation as a textured mesh with high-resolution diffusion prior by using the second sub-model in the fine stage.

### Claim 12 (depends on 10)

The system of claim 10, the operations further comprising:
refining texture and geometric details of the 3D human representation based on zooming in on body parts of the 3D human representation and applying the SDS loss.

### Claim 13 (depends on 10)

The system of claim 10, the operations further comprising:
rendering a plurality of images of the 3D human representation from a plurality of camera viewpoints.

### Claim 14 (depends on 13)

The system of claim 13, the operations further comprising:
rendering a plurality of meshes corresponding to the plurality of images from the plurality of camera viewpoints; and
utilizing the plurality of meshes as conditions for the iterative optimization to ensure view-consistency and to enable accurate pose control of the created 3D avatar.

### Claim 15 (depends on 10)

The system of claim 10, the operations further comprising:
receiving a reference image indicative of a particular style; and
creating the 3D avatar by applying the SDS loss based on the input text, the information indicative of the poses, and the reference image, wherein the created 3D avatar has the particular style.

### Claim 16

A non-transitory computer-readable storage medium, storing computer-readable instructions that upon execution by a processor cause the processor to implement operations comprising:
generating a 3D human representation with a pre-defined pose in a canonical space by adopting a coarse-to-fine pipeline, wherein the 3D human representation corresponds to a 3D avatar to be created based on input text;
deforming the 3D human representation to a target pose defined by a Skinned Multi-Person Linear (SMPL) parameter in a deformed space; and
creating the 3D avatar by implementing iterative optimization of the 3D human representation in the canonical space and the deformed space by applying a Score Distillation Sampling (SDS) loss based on the input text and information indicative of poses, wherein the poses comprise the pre-defined pose and the target pose.

### Claim 17 (depends on 16)

The non-transitory computer-readable storage medium of claim 16, wherein the coarse-to-fine pipeline comprises a coarse stage and a fine stage, and wherein the generating a 3D human representation comprises:
learning a static human representation with the pre-defined pose by leveraging low-resolution diffusion prior using a first sub-model in the coarse stage;
initializing a second sub-model with a mesh resulted from the coarse stage; and
modeling the 3D human representation as a textured mesh with high-resolution diffusion prior by using the second sub-model in the fine stage.

### Claim 18 (depends on 16)

The non-transitory computer-readable storage medium of claim 16, the operations further comprising:
refining texture and geometric details of the 3D human representation based on zooming in on body parts of the 3D human representation and applying the SDS loss.

### Claim 19 (depends on 16)

The non-transitory computer-readable storage medium of claim 16, the operations further comprising:
rendering a plurality of images of the 3D human representation from a plurality of camera viewpoints.

### Claim 20 (depends on 19)

The non-transitory computer-readable storage medium of claim 19, the operations further comprising:
rendering a plurality of meshes corresponding to the plurality of images from the plurality of camera viewpoints; and
utilizing the plurality of meshes as conditions for the iterative optimization to ensure view-consistency and to enable accurate pose control of the created 3D avatar.
