Multi-subject semantic attribute binding image generation method based on diffusion model

By optimizing the input and cross-attention mechanism of the diffusion model and combining layout and reference images, the semantic alignment problem of the diffusion model in multi-object, multi-attribute scenarios is solved, achieving efficient semantic attribute binding and spatial control, and improving the accuracy and consistency of generated images.

CN121685718APending Publication Date: 2026-03-17FOSHAN UNIVERSITY
0 Cites 0 Cited by

Patent Information

Application Number
CN202511669797.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-14
Publication Date
2026-03-17

AI Technical Summary

Technical Problem

Existing diffusion models suffer from semantic interleaving, attribute mismatch, and layout disorder in complex text-to-image generation tasks involving multiple objects and attributes, leading to semantic alignment failures and limiting their application effectiveness in complex scenarios.

Method used

By introducing layout and reference images, and optimizing the input, denoising process, and cross-attention mechanism of the diffusion model, text sentences are segmented into independent clauses for encoding. Combining spatial layout and image features, the cross-attention mechanism is improved to achieve independent generation of multi-subject semantics.

Benefits of technology

It improves semantic consistency and visual quality in multi-subject, multi-attribute scenarios, reduces semantic mixing and feature interference, and improves the accuracy and consistency of generated images. It requires no additional training and has low computational overhead.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a multi-subject semantic attribute binding image generation method based on a diffusion model. The method comprises the steps of improving independent coding of a single sentence into segmentation of the sentence into a plurality of sub-sentences and independent coding of the sub-sentences; improving a denoising process and improving a cross attention mechanism; clear semantic representation is obtained by independently coding clauses, a denoising process is divided into a reconstruction branch and a generation branch, intermediate potential representation is spliced and fused to improve consistency and visual quality, and meanwhile, independent attention guidance is realized by combining an improved cross attention mechanism and mask division, so that feature aliasing is avoided, and the accuracy and the reliability of the system are improved. And the local precision and the space decoupling capability are enhanced. According to the multi-entity semantic modeling and image-text alignment method, a finer and more stable solution is provided while high efficiency and universality are ensured, and the potential in multi-entity semantic modeling and image-text alignment is shown.
Need to check novelty before this filing date? Find Prior Art