Reference-Guided Image Generation Through Semantic Feature Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current image generation technologies, such as GAN and VAE, struggle to generate images according to user-specific needs, particularly in special scenarios requiring images with similar content, theme, and style.

Innovation Solution

An image generation method that involves acquiring a reference image and noise image, extracting a reference feature vector, and performing denoising processing using a series of diffusion models to produce a target image with partially same high-level semantic information as the reference image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional image generation methods (GAN, VAE) are used, then image generation capability is provided, but the ability to generate images according to user-specific needs and maintain consistent high-level semantic information is insufficient

Engineering Contradiction:
Improveimage generation adaptability to user needsVSAvoidcontrol precision of high-level semantic information
Core Design Contradiction:
Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent introduces a reference image as an intermediary element that mediates between user requirements and the image generation process. The reference image encodes the desired high-level semantic information (content, theme, style) and guides the generation of target images, ensuring consistency with user-specific needs while maintaining control over semantic features.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the reference image into a reference feature vector through feature extraction, changing the parameter representation from pixel space to feature space. This parameter transformation enables precise control over high-level semantic information while maintaining the ability to generate diverse target images by varying noise inputs.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If diffusion models are used for denoising processing, then image quality and semantic consistency are improved, but computational complexity increases

Engineering Contradiction:
Improveimage generation precisionVSAvoidmodel complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary feature extraction to obtain a reference feature vector from the reference image before the denoising process. This preliminary action prepares the guidance information in advance, allowing the diffusion model to efficiently utilize pre-extracted semantic features during denoising, thereby improving image generation precision while managing computational complexity through staged processing.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If multiple denoising operations are performed to maintain semantic consistency, then image generation effectiveness is enhanced, but processing time increases

Engineering Contradiction:
Improveimage generation effectivenessVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent maintains continuous guidance from the reference feature vector throughout the multi-step denoising process. Instead of performing discrete, independent denoising operations, the reference feature vector provides continuous semantic guidance across all denoising steps, ensuring semantic consistency while optimizing the balance between generation effectiveness and processing time through coherent iterative refinement.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4641514A1Image generation method and apparatus, computer readable storage medium, electronic device and computer program product
Publication Date: 2025.10.29 BEIJING ZITIAO NETWORK TECH CO LTD
  • EP4641514A1 patent drawingFigure 1~2
  • EP4641514A1 patent drawingFigure 3
  • EP4641514A1 patent drawingFigure 4~5

AI summary

Provided in the present disclosure are an image generation method and apparatus, a computer readable storage medium, an electronic device and a computer program product. The method comprises: acquiring a reference image and a first noise image; acquiring a reference feature vector corresponding to the reference image; performing denoising processing on the first noise image on the basis of the reference feature vector, so as to obtain a target image which is a variation of the reference image, the target image and the reference image having partially same high-level semantic information. The embodiments of the present disclosure improve the image generation effect, and provide richer image generation paths for users.