Visual Content Generation With Thinking-Stage Large Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing large models generate visual content with poor quality that fails to meet user requirements effectively.

Innovation Solution

Incorporate a thinking stage in the large model process to generate thinking process information based on user instructions, utilizing multimodal large models, and employ training methods like autoregressive training and reinforcement learning to enhance the accuracy of visual content generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional large models are used to generate visual content directly from instructions, then the generation process is simple and fast, but the quality and accuracy of generated visual content is poor

Engineering Contradiction:

Improvequality of generated visual contentVSAvoidcomplexity of generation process

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent segments the visual content generation process into distinct stages: instruction understanding, thinking process generation, and visual content generation. By introducing an intermediate thinking process stage that separates reasoning from generation, the model can systematically analyze instructions and plan before producing visual content, thereby improving quality without requiring complete process redesign

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies preliminary action by generating thinking process information before actual visual content generation. The model first produces reasoning traces, analysis, and planning information that guide subsequent generation steps, ensuring that the final visual output is based on thorough preliminary consideration of the instruction

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If thinking process information is generated before visual content, then the accuracy of visual content generation is improved, but the generation time and computational resources increase

Engineering Contradiction:

Improveaccuracy of visual content generationVSAvoidgeneration time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent maintains continuity of useful action by making the thinking process generation and visual content generation an integrated, continuous process rather than separate discrete steps. The thinking process tokens are generated autoregressively and immediately fed into the visual generation pipeline, eliminating idle time and ensuring that the computational work flows continuously from reasoning to generation

Inventive Principle:

Principle #20Continuity of useful action

Data Source

PatentUS20260011045A1Large model-based visual content generation and target large model training methods

Publication Date: 2026.01.08 BEIJING BAIDU NETCOM SCI & TECH CO LTD

AI summary

Large model-based visual content generation and target large model training methods, relating to artificial intelligence fields such as deep learning, a large model, computer vision and natural language processing, are provided. A large model-based visual content generation method may include: obtaining target instruction information; inputting the target instruction information into a target large model to obtain and output corresponding target result information, where the target result information includes target visual content, the target result information is generated by the target large model according to target thinking information, and the target thinking information is thinking process information generated by the target large model for the target instruction information.