Prompt2Diagram: Transforming Natural Language into Visual

Diagrams Using Advanced NLP and LLM

M. S. Minu, Ashwin Babu, Senthil R. and Tanveer Ahmed S. S.

Department of Computer Science Engineering, SRM Institute of Science & Technology, Ramapuram, Chennai, Tamil Nadu,

India

Keywords: Large Language Models, Automated Diagram Generation, AI‑driven Visualization, NLP for Creating

Diagrams, Intelligent System Modeling, Semantic Interpretation, Prompt‑Based Diagram Generation, Natural

Language Processing, Visual Documentation Automation, AI‑driven Diagram Synthesis.

Abstract: Creating diagrams (such as flow charts, sequence diagrams, Gantt charts, class diagrams, state diagrams, user

journey diagrams) is time-consuming, cognitively demanding, and typically manual endeavors. Although

traditional techniques use either manual implementation or rule-based automation, they can be

computationally inefficient at times, and at best, not widely accessible to non-technical users. Another

problem in this domain is the difficult task of producing precise and comprehensive visual documentation

(including user-required diagrams) for communications and decision making in software development, system

design, and project management. In this context, we propose Prompt2Diagram, a novel system which

automates the generation of diagrams from natural-language prompts. Drawing on the semantics of LLms

(larger than human-readable models), prompt2Diagram translationally interprets user requirements, that are

expressed in plain English, and converts them into precise and contextually pertinent diagrams.

Prompt2Diagram has been developed by applying the advanced semantic knowledge of large language

models to the problem domain, and this approach provides efficient, semantically rich, and adaptable diagram

generation mechanisms. By automation and semantic generalisation, Prompt2Diagram can be perceived as an

artificial intelligence (AI)-driven paradigm to improve the efficiency of manual diagram creation tasks while

decreasing the cognitive burden associated with manual diagram creation.

1 INTRODUCTION

The process of visual diagram generation is really

important in various domains, such as software

development, system design, and project

management. Effective visual diagram representation

of information enables teams to understand complex

structures, communicate ideas efficiently, and ensure

smooth collaboration. Manual methods of creating

flowcharts, sequence diagrams, Gantt charts, class

diagrams, state diagrams, and user journey diagrams

includes several challenges. Such classic designs are

generally slow, as users must organize elements

regime, layout adjustment and also the balance.

Moreover, it also requires specialization; to make the

process harder for people who have not dealt with

such tools or schemas before. This is compounded by

the cognitive effort needed to process abstract ideas

and distill them into organized forms, resulting in

numerous inconsistencies, mistakes, and time waste.

The updated versions of diagramming software have

reinforced the need for fast-paced and highly

productive interactions at work environments.

To deal with these problems, Prompt2Diagram

introduces a novel AI-driven approach that leverages

Large Language Models (LLMs) to automate the

generation of diagrams directly from natural language

descriptions. This groundbreaking tool enables users

to express their ideas, workflows, or system structures

in plain English, eliminating the need for manual

structuring and diagramming expertise. Unlike the

old methods that require idea on how to use specific

tools, syntax, and formatting rules, Prompt2Diagram

dynamically translates textual prompts into precise,

contextually relevant visual outputs. By harnessing

the power of LLMs' advanced semantic

understanding, this tool ensures accuracy,

adaptability, and efficiency in the diagram creation

Minu, M. S., Babu, A., R., S. and S. S., T. A.

Prompt2Diagram: Transforming Natural Language into Visual Diagrams Using Advanced NLP and LLM.

DOI: 10.5220/0013909600004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

161-169

ISBN: 978-989-758-777-1

161

process. It democratizes access to structured

visualization, making it accessible to both technical

and non-technical users, empowering them to

generate high-quality diagrams effortlessly.

The increasing difficulty in using the modern

software models, workflows, and system interactions

needs a more intelligent and intuitive approach to

visual documentation. Prompt2Diagram effectively

bridges the gap between conceptualization and

structured representation, significantly enhancing

clarity, reducing cognitive load, and fostering

seamless collaboration across teams. By making the

process of diagram generation automated, it saves a

lot of time and effort and also increases productivity,

reduces errors, and improves decision-making

through clearer visual communication. This study

discovers the transformative capabilities of LLM-

powered diagramming, demonstrating its potential to

revolutionize visual documentation by offering an

innovative, user-friendly, and intelligent solution.

With Prompt2Diagram, the future of diagramming is

no longer constrained by manual effort and technical

expertise but is instead driven by the power of AI,

enabling seamless, precise, and highly effective

visual representation of ideas.

2 LITERATURE REVIEW

Using large scale scaling models (LLMS) to automate

diagram generation is a notable advance in visual

representation, addressing challenges related to

cognitive effort, accessibility and efficiency in a

variety of fields. The shift from traditional rule-based

diagrams to automated solutions is gaining increasing

attention in the research community. As a result,

many studies have investigated various aspects of this

shift, demonstrating how automation of LLM control

transforms visual documents. This progress has led to

applications being found in a variety of areas,

including software engineering, industrial

automation, risk assessment, geospatial mapping, and

cybersecurity.

A typical AI-driven Prompt2Diagram System

follows a structured approach, starting with user input

in the form of a natural language prompt. This input

is processed by an LLM, which interprets the request

and determines the appropriate diagram type based on

contextual understanding. As illustrated in Figure 1,

the system generates various diagram types, such as

flowcharts, sequence diagrams, Gantt charts, class

diagrams, state diagrams, and user journey diagrams,

each serving a specific purpose in software

development, system design, and project

management.

Fan et al. 2025 have set the stage for a profound

research study that explains the ineffectiveness of

standard diagramming methods, that are based solely

on the human effort and the use of rules to automate

the process. The study particularly emphasizes how

the cognitive load is significantly amplified according

to the requirements to carefully place the constituent

parts and consistent syntactic rules. To tackle these

issues, they propose AI-facilitated automation,

including the use of knowledge graphs and language

models (LLMs) as well as material extrusion. The

strategy allows LLMs to cooperate with knowledge

graphs to then construct visual representations, and it

can adapt them over time, so human effort becomes

less necessary for the process of automation.

Additionally, the Sun et al. 2025. document

provides the topic of LLMs in intelligent risk

assessment, for example in the coal mine safety

domain. They show on the example of text

visualization using AI how to convey the content of a

long, complicated report in a graphic form that would

be better structured and help the reader to avoid safety

problems. Through the provision of automatic

diagram production along with the risk analysis

models, it was elucidated how AI could make the

safety monitoring systems a fully-fledged part using

visualizing tools in real-time and AI assistance.

Undoubtedly, among the primary difficulties in

computer-generated diagram creation is teaching

language models to pick up the meaning of utterances

in natural language and convert them into logically

consistent diagrams. Pan et al. 2025 resolve this

peculiar task through a Graph-RAG-centric method

of diagram optimization, showing the possibility of

AI to author elaborate industrial automation

diagrams. Through their work it is clearly seen how

the comprehension of the context and the mapping of

the semantics are of great importance for the AI-

based visualization that make sure that the diagrams

are correctly reflecting the implemented business

processes.

The authors Yhdego et al. 2025 have pushed the

boundaries of defect identification in manufacturing

using the technology they developed and also A.I.

They introduced a zero-shot learning-based LBD

system that is coupled with the AI-driven ontology

generation and graphical sketching of defects. They

have used the knowledge graphs approach to improve

the accuracy of defect finding in the setting of visual

data, which is also generated by Artificial Intelligence

to match industrial standards. The work is a concrete

example of a very useful application of AI in visual

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

162

diagramming for quality control and process

optimization in various industries.

Omar et al. 2025, continue the discussion of the

influence of knowledge graphs on improving AI

generated diagram creation even more thoroughly by

taking knowledge graphs to come up with the

benchmark for a conversation as their subject. The

authors' (research attempts to utilize LLMs with

retrieval-augmentation to attain a definitive edge in

the contextuality of the AI-generated diagrams,

revealing how the structured knowledge is central to

the coherence of the diagrams. Similar to the way,

Sahadevan et al. 2025, utilize the technology of

knowledge expansion to quickly visualize the design

concept in the early developmental stage of AI to

enable the collaborative work of AI and humans to

come up with solutions for the problems in a more

breadth-first manner.

Furthermore, LLM-powered automation has

made a huge leap in creative fields. In their work, Liu

et al. 2025 delve into the application of AI technology

in game prototyping. The paper shows how LLMs are

capable of card game design and the creation of a flow

diagram through autonomous work. This research

provides a clear representation of how to reach a

structured visualization of the game more quickly by

using AI thus making conceptualization easy and less

time-consuming. Expansion of this topic is achieved

by Acazzi et al. 2025 through emphasizing how AI-

powered querying functions can evolve to generate

knowledge graph visualizations automatically, thus

enabling data researchers and enterprises a more

convenient way to access the data.

Wang et al. 2025 focus on the use of AI

diagramming for geospatial analysis where an

intelligent mapping framework is presented by the

researchers. The framework includes LLMs with

knowledge graphs to create maps automatically.

Consequently, the mechanism of the visualization of

geographic text becomes with the help of this tool,

which also opens the gate to various other

applications such as environmental monitoring, urban

planning, and GIS., etc. The utilized system is

extremely useful in that it converts the geographical

data of a textual nature into visual maps in a fully

automatic way.

Yin et al. 2025 go even further in automating the

process. They develop a self-differentiating LLM

workflow that kills manual prompting. Using the

same method, which allows a fully automated

diagram creation, AI changes the results constantly

by learning methods that are based on iterative

mechanisms of learning.

He et al. 2025 researched the potential of LLMs

in improving graph-to-text generation by focusing on

refining AI-generated knowledge graphs as well as

their visual representations. The themes of cohesion,

flexibility, and iterative feedback emerge in their

work as crucial to the secure and contextually valid

transformation of the AI-generated drawings. The

research of Wang et al. 2025 offers a practical

solution to automation by doing the AI-generated

debugging flow diagrams of the user. Their system,

consisted of LangGraph, visualizes the bug-fixing

processes, thus automatically assists the debugging of

complex software systems.

The idea of AI-driven diagramming as a means of

policy enforcement was taken to the next level by Wu

et al. 2025, using rule-based AI recommendations to

create diagrams that would make sure the compliance

tracking process is automated. Indeed, by the simple

visual representation of the rules, their system not

only increases the transparency in the regulation but

also helps in regulatory adherence. Another very

interesting study in the sphere of AI-driven testing

presented by Kong et al. 2025 where they were able

to show how LLMs can be utilized to create a security

diagram for a multi-agent cybersecurity scenario is a

very interesting study. The authors' case describes

how AI-generated diagrams can support vulnerability

analysis, security auditing, and threat modeling.

Stennett et al. 2025 propose to end the discussion

by demonstrating a case where LLMs were used to

automate API testing documentation using a visual

workflow. Their research indeed is an example of the

use of AI in API testing and the importance of

automation in software development as that results in

structured diagrams of API testing and an increase in

the clearly show and communicate aspects of the

development teams that further lead to efficiency and

quality products.

Clearly, these separate researches highlight once

again the great potential of LLMs in the automation

of diagram generation. Numerous are the uses of

artificial intelligence to assist the generation of

knowledge graphs, to draw a workflow diagram, to

set up a security model, and to create video games.

With AI visualization being the main factor, the

whole process of visualizing complex data is facing a

whole new world of things that can be shown and

things that can be explained. In the field of automatic

diagram generation, as AI and LLMs technologies

advance further, a radical change is expected not only

in the traditional areas of real-time accessibility and

quality decision-making but also in new areas of

process effectiveness where before we had no control.

The continuous development of AI-supported

visualization techniques will not only mechanically

Prompt2Diagram: Transforming Natural Language into Visual Diagrams Using Advanced NLP and LLM

163

enhance the existing order of things but will also be

able to make new human-AI associations.

3 EXISTING SYSTEM

The existing system for making diagrams mostly

depends on traditional diagramming tools like

Microsoft Visio, Lucid Chart and similar tools, which

need the users to create visual representations

manually by selecting shapes, defining connections,

and arranging things in a structured format. Although

they have a wide variety of ready-to-use templates

and the user can use the automation features, the

changed mode is still using up a lot of time and may

sometimes disturb the person. Thus, the pathway will

be that the user will have to delve into the layout that

is complex, make changes of the configurations

manually if needed and ensure the logical consistency

of the diagrams that is way more difficult in the case

of, UML diagrams, sequence charts, workflow

models, architectural diagrams. Moreover, most of

these tools need a good understanding of specific

syntaxes or diagramming conventions which brings

to the users an extra layer of difficulty in case they are

not good at developing structured visual

documentation.

Diagrams such as class diagrams, state diagrams,

and user journey maps are typically represented by

notations that are known to a set of people who have

received a certain form of education. This makes it

difficult for persons who have not had training or are

not conversant with the tools to access the drawing

without any help. In addition, those who are already

experts face the burden of ensuring correctness and

consistency across multiple diagrams, which still is a

major challenge. Minimal modifications on one spot

require deformation of other interconnected parts,

resulting in inefficiency and a heavy load of work.

Oftentimes, offline tools are insufficient in terms of

collaborative aspects; hence, the tasks of the team are

made more complex due to this fact. Most current AI-

assisted workflows have no provision for

diagramming. The majority of the solutions are

designed in such a way that they only work in

isolation and hence, the manual entry of data is the

only option, rather than diagrams being created

automatically from the understanding. As a result, the

gap between the conception of an idea and its

execution stretches the time it takes to carry out the

process, making it harder for the company to adjust to

more fast-changing customer needs, and making it

more cumbersome for developers to adjust as quickly

as required. The existing method of visual

documentation is overall still a system marked by

inefficiency as it highly depends on human

assistance, knowledge and thus requires constant

updates to sustain accuracy and consistency as it is

now. The demand for an intellectual, automated

solution that eases the diagramming process, reduces

mental work, and enforces collaboration is very high,

this is because the number of modern projects is

getting high.

4 PROPOSED SYSTEM

The new system, Prompt2Diagram, is primarily using

an AI-based approach aimed at automating the

process of creating orderly visual representations

with the help of Large Language Models (LLMs).

The key difference with traditional diagramming

tools that rely on manual structuring or rule-based

automation is that Prompt2Diagram applies the

natural language processing (NLP) to the system for

the diagrams to be immediately generated directly

from the textual descriptions. People can verbally

describe and the system translates these requirements

into automatically structured diagrams like

swimlanes, sequence diagrams, Gantt charts, class

diagrams, state diagrams, and user journey diagrams

in a contextual way. At the heart of the system is an

advanced LLM that uses the idea of semantic

relationships, logical dependencies, and context

nuances to understand the users' questions. With the

support of Figure 1, the AI engine is examined as

dynamically overcoming the difficulties of

interpreting entangled queries, identifying well-

defined entities and relationships, as well as finally

delivering a corresponding visual output in a

standardized diagramming format. The visualization

of the model involves a rectification process depicted

by a validation mechanism that corrects the generated

diagrams on the basis of feedback, thereby promoting

the iterative process. The architecture of the platform

is such that it has quasi-perfection because the clients

are served without prejudice to those who are not

necessarily skilled in the field of technology or

technical matters to be exact. Furthermore,

Prompt2Diagram helps in enhancing efficiency and

collaboration by working together with cloud-based

platforms that allow people to edit, share, and control

versions in real-time. Additionally, the AI-powered

robotic process significantly minimized the cognitive

burden on humans associated with manual

diagramming which in return has led to increased

productivity and a reduction in the number of human

errors. Between the stage of conception and the stage

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

164

of structured presentation, the representation of the

model's process brings a big difference in the field of

system development, system design, and project

management, which has been largely eased by

Prompt2Diagram.

5 IMPLEMENTATIONS

When aligning the implementation process with the

AI-driven automated diagram generation, the

attention changes into setting up a strong technical

foundation for a flow of unproblematic conversion of

textual prompts to a visual idea of a given structure.

This part is for specifying hardware, software,

algorithmic structure, and giving an evaluation of the

system's benefits and drawbacks.

5.1 Hardware and Software

Specifications

For the proper functioning of Prompt2Diagram, it is

essential to meet specific hardware requirements. The

operating system will need a processor like the AMD

Ryzen 5 5600H or an Intel i5/i7 equivalent having

multi-threading support. A minimum of 8GB RAM

will be indispensable for basic operation, but still,

16GB or more is a better choice for improving

performance and dealing with larger AI-based

computations. Besides, for rendering support, we

would advise a dedicated NVIDIA RTX Series GPU,

or in the AMD series, an integrated AMD Radeon

Graphics for speeding up the AI-generated diagrams

processing.

For Prompt2Diagram to be a viable software

stack, it incorporates several technologies that should

guarantee a hitch-free development and deployment.

The system is actually for the Ubuntu 22.04.5 LTS,

while at the same time, it is also compatible with

Windows 10/11 and macOS. Python 3.8+ would be

the primary programming language, with a number of

NLP and AI frameworks like OpenAI GPT API,

Hugging Face Transformers, and SpaCy for natural

language understanding being utilized. In the process

of image projection, the Mermaid.js is selected as the

tool to convert structured data into graphical form. It

becomes the backend when the Python is put into use

and there is also efficient data processing and

management of the API. In writing of the code, the

tools VS Code or PyCharm, Postman for API testing,

and Docker for containerization are used, which

guarantee scalability and platform independence.

5.2 Algorithmic Framework

The Prompt2Diagram system is fitted with the

Natural Language Processing (NLP) engine using AI,

which transforms its content from the written form to

structured diagrams. The first step is the processing

of prompts and the subsequent recognition of

semantic content with the system finding and then

extracting a user's query. At this level, the system is

also responsible for parsing the objects, the

relationships and the context of the user's input. The

next step involves the creation of concrete diagrams

and their mapping. During this step, the data that has

been separated is sorted according to appropriate

types of diagrams such as flowcharts, sequence

diagrams, and class diagrams. The final phase of the

generation of the diagrams and their visualization is

where the conversion of the data to the graphical form

of the Mermaid.js library takes place, which itself is

complemented with a real-time response. The

validation and optimization step ensure that the

produced diagrams are both correct and logically

coherent, thus allowing for the user's feedback to be

incorporated into the diagrams.

5.3 Advantages

The Prompt2Diagram system has a number of

benefits, especially in that it automates the diagram

making process and optimizes it as well. The software

developers, project managers, and system designers

are the biggest beneficiaries as it not only increases

their productivity but also saves them a lot of effort

they initially used to put while structuring the

diagrams manually. The NLP-based approach, on the

other hand, is the innovative text-to-diagram

conversion, which makes the system accessible to

non-technical users. Also, it is cross-platform

compatible and can run on different operating

systems, therefore, highly portable. The real-time

nature of rendering and editing, which are the main

features of the system, have the potential to transform

the diagram sharing and modifying processes

completely, thus making the latter more efficient.

Furthermore, the AI-powered validation

mechanisms that have been adopted play a big role in

reducing human errors, as they can categorically

certify that the final diagrams are logically accurate

and structurally integral.

6 5.4 DISADVANTAGES

Although Prompt2Diagram has a variety of

Prompt2Diagram: Transforming Natural Language into Visual Diagrams Using Advanced NLP and LLM

165

advantages, it is by no means a perfect system. The

initial modeling and fine-tuning work need a large

number of computational resources, which will

eventually make both the setup and deployment

resource consuming. When the user input is unclear,

the deep-learning-based system might get it wrong

and human intervention is needed to correct the

situation. When it comes to the problem of large

enterprise-level datasets, it is quite possible that

scaling issues might occur as a result, which will in

turn lead the way to more hardware and infrastructure

setup. The alignment of diagramming standards

across industries may lead to complex issues, and this

can be a barrier to the success of the system among

traditional users who may lack the necessary outreach

and training for exploiting the system's capabilities.

Overcoming these challenges requires regular

improvements in AI models, step-by-step

refinements, and user feedback-driven enhancements.

7 ARCHITECTURES

Figure 1: Architecture diagram of the proposed model.

The Prompt2Diagram architecture is a new advanced

system that allows a user to get a structured diagram

from natural language input automatically. Diagram

generation is performed by the Large Language

Model (LLM). The procedure is commenced when a

user gives a natural language prompt that refers to the

kind of diagram such as, for example, a flowchart, a

sequence diagram, or a Gantt chart. This type of input

is firstly handled by the Prompt2Diagram system, it

not only interprets the user request but also enriches

it with metadata and thus gives the main keywords

and the context the user is in. The cleaned prompt is

then given to the Large Language Model (LLM). The

LLM is the central and most critical part of the tool

that generates the new diagrams by mainly analyzing

the user's request. It thus tries to find out which one

of the many different types of diagrams not only suit

the users but also really meet their intentions. As a

result of processing the input, an LLM not only finds

an answer to the user but also a drawing (ensuring

similarity to the real line of thinking and the structure

of the question), and thereby it generates the desired

diagram. The tool is equipped with a variety of

diagrams for all types and fields, such as flowcharts,

sequence diagrams, Gantt charts, class diagrams, etc.

Each diagram type serves specific needs for each use

- flowcharts describe the execution of processes,

sequence diagrams reflect the interactions in software

development, Gantt charts represent plans for

projects, etc. Class diagrams specify the

characteristics and features of object-oriented system

structures, state diagrams portray the shift from one

state of a system to another state, user journey

diagrams show users views on a project in detail.

The project is specifically developed to be capable

of implementing the traditional manual way by

exploiting the automatic diagram making technology.

The time dropped from the traditional manual way

with the feature of an efficient process of diagram

creation not only is fast but is also beneficial to the

creators in that they can comfortably concentrate on

high-level conceptualization and refrain from the

rigors of the technical drawing. Once the tasks are

automated in this way, productivity is boosted,

unforced errors are reduced, and visual

documentation is made more approachable to a wider

audience including the software developers, the

project managers, the system designers, and the

business analysts. Outstanding visual content is

produced by this directive of changing of Hand Open,

where, in effect, the users execute the same operation

as was mentioned in the previous point, and then,

through the generation of not somehow with the exact

data but on the contrary, standard and general text-

based, they turn it into visual information. The result

is that the natural language user clearly understands

the purpose and creates the diagram in a neutral

format, thus, the gap is bridged between the natural

text and the visual.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

166

8 METHODOLOGIES

The research is based on the design and release of

Prompt2Diagram, an AI-powered system that

transforms natural language inputs into

understandable visual diagrams. The process includes

four main steps:

8.1 Module 1: Natural Language

Processing (NLP) Module

This section is in charge of understanding textual

descriptions given by the user by taking out the main

concepts, connections, and contextual meanings from

the diagram so that it can be structured. It employs

transformer-based models to get information out of

input prompts and then winnow information down to

the most important diagram elements namely

processes and decision points and at the end build up

the connections between them. The deep semantic

relations of the spoken word are captured by the

model to ensure that the complex text-based inputs

are translated to the structural representation

correctly.

8.2 Module 2: Diagram Generation

Engine

The diagram generation engine converts the treated

text into more or less structured visual diagrams by

using the Mermaid.js and other graph-rendering

libraries. It identifies the diagram type according to

the source data, e.g., flowcharts, Gantt charts, and

sequence diagrams, and constructs nodes and edges

that correspond with the relationships, which were

extracted. It also arranges the diagram components

for better understanding and removes the layout

ambiguity. As a result, a user can conceive new ideas

unchallenged without structuring them manually.

8.3 Module 3: User Interface (UI)

Module

The system is interacted with without any friction by

a web interface that is easily used by the customers.

Users can put in the verbal form of their ideas, and the

application will instantly display them in visuals. The

interface includes a variety of options that ensure the

flexibility of creating the diagrams such as changing

node labels, connections, and layout styles. Not only

that, but the users are able to save the diagrams in

different files like PNG, SVG, and PDF, so that they

can be used in the professional world in different

ways.

8.4 Module 4: Error Handling &

Optimization Module

For high-quality and logically precise diagram,

one could use this module that will combine the

checking and cleaning steps. The checking part

entails the syntax validation to point out the cases of

incomplete or ambiguous input prompts, and the

correction step is all about the semantic correction to

improve the text's understandability. Furthermore, it

handles the optimization of the graph layout by

avoiding the elements to cover each other and by

taking care of their contour. This module also

automates the process of finding errors and cleaning

up the layout of the generated diagrams, which in turn

result in error-free and user-friendly diagrams.

This method gives a logical sequence in the

construction and implementation of

Prompt2Diagram, so that the product is the most

competent with the least interference from the user,

turning text explanations into professional and

beautiful diagrams.

9 RESULTS

The performance metrics of this system are presented

in terms of graphs that are created to measure

performance.

Figure 2: Accuracy Comparison of Diagrams Generated by

Prompt2Diagram vs. Manual Methods.

Prompt2Diagram: Transforming Natural Language into Visual Diagrams Using Advanced NLP and LLM

167

Figure 3: Reduction in Diagram Creation Time – Manual

vs. Prompt2Diagram.

Figure 2 shows a comparative analysis of diagram

accuracy between manually created diagrams and the

ones made with Prompt2Diagram. The evaluation of

correctness in the structure, relationships, and layout,

which is the basis for measuring accuracy, is done by

the experts. The outcomes show that the usage of

Prompt2Diagram led to a notably higher accuracy,

which in turn means that the inconsistencies were

curbed that are usually found in manual

diagramming.

According to Figure 3, the tool can reduce the

time spent on creating diagrams. A visual in the shape

of a stickman is the representation of a comparison

showing the difference between the average time that

was used by the participants in two groups to create

diagrams by hand and the time that was needed when

using Prompt2Diagram. The results are evidence of a

major drop in the time required for the creation of the

diagram. It also shows that automation is of real use

by a significant process of improving and reducing

the user effort and cognitive difficulty to a minimum.

Figure 4: Proportion of Errors – Manual vs. AI-Generated

Diagrams.

Lastly, Figure 4 reveals the difference in error ratio

between manual diagrams and those produced by AI.

The findings prove that Prompt2Diagram is mainly

responsible for the decrease in the number of

inaccuracies, especially in three main categories:

wrong connections, unfinished components, and

inconsistent formats. The validation feature that

follows the system's operation automatically

guarantees the more precise nature of the drawings

made by the system. In this way, it becomes

unnecessary to do editing after the diagrams are

created.

10 DISCUSSION

The enhanced diagram accuracy confirms integrating

the Prompt2Diagram AI system would drastically

reduce errors and inconsistencies inherent in manual

diagramming. Utilizing Large Language Models

(LLMs) for semantic interpretation guarantees

generated diagrams possess structural soundness and

conform to established standards. The considerable

reduction in diagramming errors, proven by the

results, shows the power of automated natural

language processing in converting textual

descriptions into accurate visual depictions.

Furthermore, the reduced time for diagram creation

reinforces the system's efficiency, enabling users to

produce intricate diagrams far faster than traditional

manual methods. This time saving not only lessens

mental strain but also boosts overall productivity,

democratizing diagramming for professionals in

sectors like software development, project

management, and system design. Finally, the marked

rise in user adoption rates after implementation

indicates the system's intuitive design and ease of use.

The accessibility component of Prompt2Diagram

is very important in the adoption of its rapidly

increasing end-users, especially the non-technical.

The feature of generating diagrams from the user's

English wording not only takes down the entry-level

but also allows for a wider range of people to be

involved with the diagramming tools without the need

for specialized knowledge. High usage of such

advanced diagram categories, for example, sequence

diagrams and class diagrams, and the lower usage of

the trivial ones like flowcharts also reveal that

Prompt2Diagram can be used for problems of

different abstraction levels thus proving its

effectiveness in the differentiation of the application.

In addition to the tool's ability to handle complex

diagram types that are commonly used in the industry,

this feature further enhances the tool's adaptability for

different sectors. The effective time it takes to process

the system across various types of diagrams is another

proof of its real-world usability.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

168

The automation of diagram structuring, layout

optimization, and error handling guarantees that the

user gets the best outputs without delay-- and all that

is accomplished without the irritating manual

corrections which are usually needed in the traditional

tools.

In summary, the research outcomes and pictorial

illustrations validate Prompt2Diagram as a proper

vehicle for making diagrams not only more accurate

and efficient but also more accessible. This is a real

breakthrough that allows the human language to be

directly transposed into the visual form and makes the

tool of great capability for all professionals and teams

who want to simplify their documentation workflow.

The growth in the number of satisfied users and the

identified decrease in the number of errors also prove

the practicality of AI in diagram creation and hence

provide strong evidence that AI-based visual

documentation workflows can drive the change

towards modern practices.

11 CONCLUSIONS

The research study reveals a plan, guiding the

scheme, Prompt2Diagram, that depends on Large

Language Models and Generative AI to remake NLP

descriptions into chips of diagrams, which in other

words, will lead to a qualitative change of the visual

documentation's mechanism by way of effectiveness,

accessibility, and accuracy as such. With the process

of making the drawing automated, Prompt2Diagram

gives a necessary instrument to professionals in the

area of software development, system design, and

project management. This LLM-based diagram

creation is subject to a thorough investigation and it

is apparent here that the system's capacity and

adaptability are being described while the system

architecture, basic modules, and user interactions are

being mentioned. NLP algorithms working hand in

hand with graph-rendering tools like Mermaid.js

enable the system to create diagrams with more

precision, thereby reducing the chances of errors and

making the steps clearer in the case of complicated

processes. The scope of the project in the future is

likely to be aimed at adding real-time collaborative

features, enhancing the error-handling process, and

embodying support for more diagram formats, so that

the experience of the user is still optimized. The study

pushes the frontiers of AI-powered visual

documentation by proposing unique and automated

ways of converting ideas into coherent visual

displays.

REFERENCES

D. Li, S. Zhang, SS Sohn, K. Hu, M. Usman, Cardiverse:

Harnessing LLMs for Novel Card Game Prototyping,

arXiv, 2025.

G. Wu, L. Hu, Y. Hu, X. Xiong, LLM4TAP: LLM-

Enhanced TAP Rule Recommendation, IEEE, 2025.

H. Fan, J. Huang, J. Xu, Y. Zhou, JYH Fuh, WF Lu, B. Li,

AutoMEX: Streamlining Material Extrusion with AI

Agents Powered by Large Language Models and

Knowledge Graphs, ScienceDirect, 2025.

H. Kong, D. Hu, J. Ge, L. Li, T. Li, B. Wu, VulnBot:

Autonomous Penetration Testing for a Multi-Agent

Collaborative Framework, arXiv, 2025.

J. He, Y. Yang, W. Long, D. Xiong, VG Basulto,

Evaluating and Improving Graph-to-Text Generation

with Large Language Models, arXiv, 2025.

J. Wang, Z. Duan, Empirical Research on Utilizing LLM-

Based Agents for Automated Bug Fixing via

LangGraph, Cambridge University Press, 2025.

L. Yin, Z. Wang, Auto-Differentiating Any LLM

Workflow: A Farewell to Manual Prompting, arXiv,

2025.

M. Arazzi, D. Ligari, S. Nicolazzo, A. Nocera, Augmented

Knowledge Graph Querying Leveraging LLMs, arXiv,

2025.

M. Wang, B. Li, Z. Wang, S. Liu, C. Liao, An Intelligent

Mapping Framework Integrating Knowledge Graphs

and LLMs, IEEE, 2025.

R. Omar, O. Mangukiya, E. Mansour, Dialogue Benchmark

Generation from Knowledge Graphs with Cost-

Effective Retrieval-Augmented LLMs, ACM Digital

Library, 2025.

T. Pan, W. Pu, L. Zhao, R. Zhou, Leveraging LLM Agents

for Automated Optimization Modeling for SASP

Problems: A Graph-RAG Based Approach, arXiv,

2025.

T. Stennett, M. Kim, S. Sinha, A. Orso, AutoRestTest: A

Tool for Automated REST API Testing Using LLMs

and MARL, arXiv, 2025.

T.O. Yhdego, H. Wang, Automated Ontology Generation

for Zero-Shot Defect Identification in Manufacturing,

ScienceDirect, 2025.

V. Sahadevan, R. Joshi, K. Borg, V. Singh, Knowledge-

Augmented Generalizer Specializer: A Framework for

Early Stage Design Exploration, ScienceDirect, 2025.

Y. Sun, Y. Han, X. Liu, Intelligent Gas Risk Assessment

and Report Generation for Coal Mines: An Innovative

Framework Based on GLM Fine-Tuning, MDPI

Electronics, 2025.

Prompt2Diagram: Transforming Natural Language into Visual Diagrams Using Advanced NLP and LLM

169