Multimodal Meme Generation Using Template Planning and MMR Retrieval

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional AI-based meme generation techniques rely on single-modal input, such as text prompts or user-provided images, which are restrictive and challenging for clearly specifying the intent, limiting the effectiveness of meme creation.

Innovation Solution

A multi-modal input approach using a large language model (LLM) and vision language model (VLM) to generate a contextual template plan, followed by recontextualization and retrieval of a final meme template using max marginal relevance (MMR) search, combined with caption text planning to create memes from both image and text inputs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If single-modal input (text prompt or user-provided image) is used for AI-based meme generation, then the system complexity is reduced, but the ability to clearly specify intent and generation effectiveness deteriorates

Engineering Contradiction:
Improveinput system complexityVSAvoidintent specification accuracy
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The patent combines multiple input modalities (text prompts and user-provided images) into a unified multi-modal input system. This allows the system to leverage both textual descriptions and visual references simultaneously, improving intent specification accuracy without significantly increasing system complexity through the use of integrated processing pipelines.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces intermediate processing components that bridge different modalities. These intermediaries process and align text and image inputs before passing them to the generation model, enabling effective multi-modal fusion while managing system complexity through modular architecture.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If multi-modal input with LLM and VLM is used for meme generation, then intent specification accuracy is improved, but system complexity increases

Engineering Contradiction:
Improveintent specification accuracyVSAvoidgeneration system complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent segments the generation system into distinct functional modules: an LLM component for processing textual intent, a VLM component for processing visual inputs, and a coordination layer that integrates their outputs. This segmentation allows each component to specialize in specific modalities, improving overall accuracy while managing complexity through modular design.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent employs multi-functional components that can handle multiple types of inputs and perform various processing tasks. The unified generation model can process both text and image inputs through the same architecture, reducing the need for separate specialized systems and thereby controlling complexity while maintaining high intent specification accuracy.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4703909A1Methods and systems for meme generation with multi-modal input and planning
Publication Date: 2026.03.04 TATA CONSULTANCY SERVICES LTD
  • EP4703909A1 patent drawingFigure 1
  • EP4703909A1 patent drawingFigure 2
  • EP4703909A1 patent drawingFigure 3A

AI summary

The disclosure relates generally to methods and systems for meme generation with multi-modal input and planning. Conventional AI-based techniques either rely on input text prompt or the user-provided image as an input to generate the meme. Such input specification styles result in restrictive for clearly specifying an intent with a single modality. The present disclosure explores a multi-modal input specification style where a user can provide the input through a text prompt along with a widely popular meme template image. According to the present disclosure, the meme generation task is defined as a combination of two sub-tasks. In the first sub-task, a meme image template is retrieved from a dataset of existing meme templates using a template planning strategy. In the second sub-task, the text caption is generated for the retrieved template conditioned on the multi-modal input provided by the user through the caption planning strategy.