Hybrid 3D Object Representations for Controllable Mesh Editing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional 3D modeling tools require significant user interaction and resource expenditure to create realistic digital content, making them inefficient and inaccessible to casual users, while existing generative AI models lack control over editing and are incompatible with conventional rendering tools.

Innovation Solution

A hybrid object construction technique using both implicit and explicit representations, where a machine-learning model, such as a generative AI, interprets natural language prompts to automatically update a neural representation of a mesh, synchronizing changes with the mesh to achieve user-requested modifications efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If conventional mesh-based modeling tools are used to create realistic digital content, then rendering quality is improved, but user interaction time and manual effort increase significantly

Engineering Contradiction:

Improverendering qualityVSAvoiduser interaction time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent uses neural representations (implicit models) as computational copies of the mesh-based explicit representations. These neural models can be trained from mesh data and used to generate and edit 3D objects without requiring direct manipulation of the original mesh geometry, thereby reducing manual editing time while maintaining rendering quality through the neural field's implicit surface definition

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces manual mechanical manipulation of mesh vertices and polygons with AI-based automated editing. The machine learning model processes natural language prompts and automatically modifies the neural representation, substituting the manual sculpting process with intelligent automation that achieves similar or better results with minimal user input

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Ease of operation

If conventional mesh-based modeling tools are used, then control over object attributes is maintained, but ease of operation decreases for casual users

Engineering Contradiction:

Improveaccessibility for casual usersVSAvoiduser interaction complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent introduces neural representations as an intermediary between the user's high-level intentions (natural language prompts) and the final mesh output. This intermediary layer absorbs the complexity of 3D geometry manipulation, allowing casual users to express their ideas through simple text without needing to understand or interact with complex mesh structures, while still achieving precise control over the final object attributes

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the complex mechanical interface of conventional modeling tools (requiring knowledge of vertices, edges, faces, normals, UV mapping) with a natural language interface. The machine learning model translates user-friendly text commands into the appropriate geometric modifications, eliminating the need for users to learn complex modeling terminology and techniques while maintaining precise control over object attributes

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If generative AI models are used for 3D object creation, then productivity is improved, but control over editing and compatibility with conventional tools deteriorates

Engineering Contradiction:

Improveediting efficiencyVSAvoidcontrol over editing

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic bidirectional conversion system that can switch between neural representation mode (for high-speed AI generation and editing) and mesh representation mode (for precise control and tool compatibility). The system dynamically selects which representation to work with based on the task requirements, allowing AI-based productivity gains when speed is needed while maintaining precise control and compatibility when required

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent uses the neural representation as an intermediary that bridges AI-generated results and conventional mesh-based tools. The neural model can be converted to a mesh format that is compatible with existing 3D software ecosystems, while also serving as an efficient intermediate representation for AI editing operations. This intermediary layer enables both high productivity through AI and full compatibility with conventional tools

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260073650A1Hybrid object constructions based on implicit and explicit representations

Publication Date: 2026.03.12 ADOBE INC

AI summary

Techniques for using hybrid object constructions based on implicit and explicit representations are described. In an example, a processing device is operable to obtain a mesh that models an exterior surface of a simulated object, generate a neural representation of the exterior surface, and receive a user input that indicates a requested modification to one or more surface regions of the mesh. The processing device is further operable to incrementally update, using a machine-learning model, the surface regions of the mesh based on incremental changes applied by the machine-learning model to the neural representation for achieving the requested modification within corresponding portions of the exterior surface of the neural representation. The processing device is further operable to output the updated mesh for use in rendering the simulated object with the requested modification.