Moodboard Augmentation With Cross-Modal Image Blending Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image curation tools for visual content creators, such as PINTEREST® and BEHANCE®, lack interactive and augmentative capabilities to stimulate creative ideation, leading to a manual and aimless exploration of design spaces.
Innovation Solution
A system and method for automated moodboard augmentation via cross-modal generative association making, utilizing a neural network to infer representative text or descriptions from selected images and generate new images based on an adjustable slider, allowing for concept blending and interpolation/extrapolation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual searching and scrolling through images is used to explore design space, then visual content creators can find inspiration, but the process is time-consuming and aimless
Solution Approach 1:
The system enables self-service by allowing the moodboard to automatically generate augmentative images based on existing selected images, eliminating the need for manual searching. The generative model autonomously creates new images that blend concepts from selected images, providing continuous design inspiration without user intervention in the search process.
Solution Approach 2:
The system performs preliminary action by pre-generating augmentative images before the user needs them. When images are added to the moodboard, the system immediately begins generating blended images in the background, so that design inspiration is already available when the user returns, rather than requiring active searching at that moment.
2Adaptability or versatility
If traditional image curation tools are used, then images can be organized and displayed, but interactive and augmentative capabilities to stimulate creativity are lacking
Solution Approach 1:
The system merges traditional image curation functionality with generative AI capabilities into a unified moodboard system. The augmentative images are seamlessly integrated with selected images in the same moodboard interface, combining organization/display features with creative generation features without requiring separate tools or complex workflows.
Solution Approach 2:
The moodboard system performs multiple functions: it organizes and displays selected images like traditional curation tools, while simultaneously generating augmentative images to stimulate creativity. The same interface handles both image display and creative generation, making the system versatile without increasing operational complexity for the user.
3Stability of the object's composition
If generative images are created with high resemblance to adjacent images, then design coherence is maintained, but design fixation may occur limiting creativity
Solution Approach 1:
The system applies dynamics by providing an adjustable slider that allows users to dynamically control the blending strength between selected and generated images. Users can shift the balance between coherence (higher resemblance) and exploration (lower resemblance) based on their current creative needs, making the system adaptable rather than fixed.
Solution Approach 2:
The system changes parameters by allowing users to adjust the resemblance parameter through the slider control. This parameter change directly affects the generative process, enabling users to explore different degrees of similarity between original and generated images, thereby controlling both coherence and creative exploration through a single parameter adjustment.
Data Source
AI summary
A method for automated moodboard augmentation via cross-modal generative association making is described. The method includes specifying, by a user, a region to augment in their digital workspace, including at least one selected image. The method also includes inferring a representative text, label, or description for the at least one selected image. The method further includes creating a basis for concept blending based on the representative text, label, or description inferred for the at least one selected image. The method also includes generating images in response to an adjustable slider, as adjusted by the user, to adjust how much the generated images should resemble directly adjacent images, including the at least one selected image.