Emulating Hand-Drawn Lines in CG Animation via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for creating artistic-looking hand-drawn line work in computer graphics (CG) animation are inadequate, as they rely on procedural rules and fail to accurately emulate the artistic feel and look of hand-drawn characters, necessitating a more robust solution that allows for adjustment and incorporation into existing CG animation.
Innovation Solution
A method and system that enable artists to draw ink lines directly onto characters, using a nudging brush to modify and project them onto 3-D geometry, with machine learning models generating training data to create realistic hand-drawn lines by combining first and second pass data, allowing for intuitive 2-D screen space adjustments that automatically translate back to 3-D space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If procedural rules and Toon Shaders are used to generate line work, then the process is automated and efficient, but the artistic quality and hand-drawn feel are compromised
Solution Approach 1:
The patent uses machine learning models to copy and emulate the patterns and characteristics of hand-drawn lines by analyzing training data from artist-created drawings. The model learns from examples and generates synthetic line work that replicates the artistic style while maintaining efficiency.
Solution Approach 2:
The system transforms 2-D screen space drawings into 3-D character space by applying parameter transformations based on character geometry and pose. This allows the line work to adapt dynamically to different character configurations while preserving the artistic appearance.
2Manufacturing precision
If ink lines are drawn directly onto 3-D characters, then the line work accurately reflects specific poses and expressions, but the workflow complexity increases
Solution Approach 1:
The patent introduces an intermediary machine learning model that mediates between the 2-D drawing space and 3-D character space. The model processes drawings in the familiar 2-D screen space and automatically transforms them to match the 3-D character geometry, simplifying the workflow while maintaining accuracy.
Solution Approach 2:
The system operates primarily in the 2-D screen space dimension where artists are comfortable working, and only transforms to 3-D space when necessary for final rendering. This dimensional approach maintains simplicity in the creation process while achieving pose-specific accuracy.
3Manufacturing precision
If artists manually adjust and refine line work, then the artistic quality improves, but the time required for production increases
Solution Approach 1:
The machine learning model provides feedback by generating initial line work based on training data, which artists can then review and adjust. This feedback mechanism allows for rapid iteration and refinement, reducing the overall time compared to manual creation while maintaining high artistic quality.
Solution Approach 2:
The system performs preliminary action by automatically generating line work using the machine learning model before artist refinement is needed. This pre-generation handles the time-consuming portions of the workflow, leaving artists to focus only on necessary adjustments and refinements.
4Manufacturing precision
If line work is created for specific character poses and expressions, then the realism is enhanced, but the adaptability to different poses is reduced
Solution Approach 1:
The machine learning model is trained on diverse training data that includes multiple poses and expressions, making it universal and capable of generating appropriate line work for various character configurations. This multi-functionality allows the same model to handle different poses without requiring separate specialized models.
Solution Approach 2:
The system dynamically adjusts line work parameters based on the specific character pose and expression by transforming 2-D drawings into 3-D space using pose-dependent parameter transformations. This allows the line work to adapt to different poses while maintaining realism for each specific configuration.
Data Source
AI summary
Methods, systems, and apparatus including: first attaching ink lines drawn directly onto characters of the CG animation; first enabling the artist to modify the ink lines using nudges to generate a drawing; first moving the drawing into a UV space and generating first pass training data; second attaching the ink lines onto the characters of the CG animation; second enabling the artist to modify the ink lines; second moving the drawing into the UV space and generating second pass training data; combining the first pass training data and the second pass training data to generate combined data; and creating and outputting a second generation machine learning using the combined data.