AI Image Generation With Retrieved Reference Images for Timeliness

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

AI image generation models often lag in generating up-to-date content due to outdated training data, leading to a timeliness issue and limited memory capacity.

Innovation Solution

An image generation method utilizing image retrieval-augmented generation (iRAG) technology, which queries reference images based on user input to align timeliness and enhance image generation, avoiding memory bottlenecks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If AI image generation models use fixed training data, then model stability is maintained, but timeliness of generated content deteriorates

Engineering Contradiction:

Improvemodel stabilityVSAvoidtimeliness of generated content

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system pre-queries and stores reference images in advance before they are needed for generation. When a generation request comes in, the system retrieves pre-stored reference images rather than training on new data, thus maintaining timeliness without compromising model stability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces reference images as an intermediary between the fixed training data and the generation task. These reference images serve as a bridge that provides up-to-date visual information without requiring retraining of the model, thus resolving the contradiction between stability and timeliness.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of time

If AI image generation models are retrained with updated data, then timeliness of generated content is improved, but computational cost and training time increase

Engineering Contradiction:

Improvetimeliness of generated contentVSAvoidcomputational cost and training time

Core Design Contradiction:

Loss of timeVSUse of energy by stationary object

Solution Approach 1:

The system extracts only the essential visual reference information needed for timely generation without performing full model retraining. By taking out just the reference images rather than retraining the entire model, the system achieves timeliness while avoiding high computational costs.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of retraining the model with updated data, the system copies relevant visual information into reference images that can be quickly retrieved. This copying approach provides up-to-date content without the energy-intensive process of model retraining.

Inventive Principle:

Principle #26Copying

3Productivity

If AI image generation models rely on internal memory, then generation speed is maintained, but memory capacity limitations arise

Engineering Contradiction:

Improvegeneration speedVSAvoidmemory capacity

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The system moves from relying solely on internal model memory to an external reference image storage dimension. By querying reference images from external storage when needed, the system overcomes internal memory capacity limitations while maintaining fast generation speeds through efficient retrieval mechanisms.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS20250378598A1Image generation method and device, intelligent agent, intelligent agent system and storage medium

Publication Date: 2025.12.11 BEIJING BAIDU NETCOM SCI & TECH CO LTD

AI summary

An image generation method includes: obtaining image generation requirement information; determining a target image generation manner according to the image generation requirement information; querying a first reference image based on the image generation requirement information; and based on the image generation requirement information and the first reference image, generating a target image using the target image generation manner.