SYSTEM AND METHOD FOR MULTILAYER DESIGN GENERATION GUIDED BY AN ANONYMOUS REGION LAYOUT

Smart overview of the Invention

A data processing system is introduced to create multilayer graphic designs using a text prompt. It involves predicting a layout with anonymous regions, each defined by a bounding box without specific content. A diffusion transformer concurrently generates multilayer image latents, including a global reference image, a background layer, and transparent foreground layers. These layers correspond to the anonymous regions and are decoded into a final multilayer graphic design, which is then displayed on a client device.

Background

AI image generation platforms have expanded, offering diverse tools for creative needs. Current diffusion design pipelines focus on visual element variation but rely on human-crafted templates for layout guidance. Multilayer transparent image generation is becoming popular, enhancing image quality and flexibility. Methods like Text2Layer and LayerDiff allow artists to create complex images, but they are limited in the number of transparent foreground layers they can generate. This system addresses the need for generating unlimited transparent foreground layers without layer-wise prompts.

Technical Approach

The system employs a processor and machine-readable medium storing instructions to execute operations for multilayer graphic design creation. It starts with receiving a text prompt, predicting a layout with anonymous regions, and generating multilayer image latents using a diffusion transformer. The vision transformer decodes these latents into the final graphic design. The process results in an output composed of a background layer and multiple transparent foreground layers, which is then displayed on a client device.

Innovations

This approach introduces an AI-based pipeline for generating anonymous region layouts guided by a global text prompt. It eliminates the need for layer-wise prompts, enabling the generation of multiple transparent foreground layers. The system uses diffusion transformer models to create these layers based on an anonymous region layout. This method allows for easy editing and scaling of the final design, addressing the limitations of existing systems that generate a limited number of layers.

Benefits

• Scalability: Supports generating a variable number of layers at different resolutions, enhancing scalability.
• Coherence: Ensures harmonious integration of elements across layers, reducing semantic gaps.
• Efficiency: Reduces computational load by focusing on one anonymous region per layer, suitable for complex designs.
• Usability: Allows users to edit elements, adjust resolutions, and vary designs at the layer level, outperforming specialized models.