PosterGen: Aesthetic-Aware Multi-Modal Paper-to-Poster Generation via Multi-Agent LLMs

Chenyu You; Jiaqi Wei; Xiang Zhang; Yiwei Xu; Zhilin Zhang

arxiv: 2508.17188 · v2 · submitted 2025-08-24 · 💻 cs.AI

PosterGen: Aesthetic-Aware Multi-Modal Paper-to-Poster Generation via Multi-Agent LLMs

Zhilin Zhang , Xiang Zhang , Jiaqi Wei , Yiwei Xu , Chenyu You This is my paper

Pith reviewed 2026-05-18 21:59 UTC · model grok-4.3

classification 💻 cs.AI

keywords multi-agent LLMsposter generationpaper to posteraesthetic designLLM agentsvisual automationconference posters

0 comments

The pith

A multi-agent LLM system turns research papers into presentation-ready posters by splitting design work among specialized agents.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes PosterGen, a framework that assigns distinct LLM agents to parse paper content, organize a storyboard, create a spatial layout, apply color and typography styles, and render the final output. This structure copies the steps professional designers follow, addressing the common shortcoming that prior paper-to-poster tools produce layouts lacking visual balance and appeal. A sympathetic reader would care because conference poster preparation consumes hours that could otherwise go to research, and the system aims to deliver results that need only light touch-ups. Evaluation relies on a vision-language model rubric that scores layout balance, readability, and aesthetic coherence, with experiments indicating content fidelity stays comparable to baselines while visual quality improves markedly.

Core claim

PosterGen deploys four collaborative agents—Parser and Curator for content extraction and organization, Layout agent for spatial mapping, Stylist agents for visual elements such as color and typography, and Renderer for final composition—to generate posters that remain semantically faithful to the source paper while satisfying core design principles.

What carries the argument

A multi-agent LLM workflow that mirrors professional poster design steps through separate agents for parsing, layout planning, styling, and rendering.

If this is right

Generated posters require only minimal human refinements before presentation.
Content fidelity to the original paper stays at least as high as existing automated approaches.
Visual design scores on layout balance, readability, and aesthetic coherence rise above those of previous methods.
The same agent分工 can be reused for other document-to-visual tasks that involve both content accuracy and design rules.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar agent divisions could shorten preparation time for conference slides or research infographics.
Replacing the current VLM evaluator with stronger vision models might further close the gap to expert-level aesthetics.
Fields that rely on frequent poster presentations would see the largest time savings if the method scales reliably across paper types.

Load-bearing premise

A vision-language model rubric can reliably judge layout balance, readability, and aesthetic coherence without direct validation against expert human designer ratings.

What would settle it

A blind comparison study in which professional designers rate PosterGen outputs against those from prior methods and find them equal or inferior on visual appeal would undermine the claim of significant design improvement.

read the original abstract

Multi-agent systems built upon large language models (LLMs) have demonstrated remarkable capabilities in tackling complex compositional tasks. In this work, we apply this paradigm to the paper-to-poster generation problem, a practical yet time-consuming process faced by researchers preparing for conferences. While recent approaches have attempted to automate this task, most neglect core design and aesthetic principles, resulting in posters that require substantial manual refinement. To address these design limitations, we propose PosterGen, a multi-agent framework that mirrors the workflow of professional poster designers. It consists of four collaborative specialized agents: (1) Parser and Curator agents extract content from the paper and organize storyboard; (2) Layout agent maps the content into a coherent spatial layout; (3) Stylist agents apply visual design elements such as color and typography; and (4) Renderer composes the final poster. Together, these agents produce posters that are both semantically grounded and visually appealing. To evaluate design quality, we introduce a vision-language model (VLM)-based rubric that measures layout balance, readability, and aesthetic coherence. Experimental results show that PosterGen consistently matches in content fidelity, and significantly outperforms existing methods in visual designs, generating posters that are presentation-ready with minimal human refinements.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

PosterGen breaks poster generation into four specialized LLM agents plus a VLM rubric, but the outperformance claims sit on unvalidated scores with no numbers or human checks shown.

read the letter

PosterGen applies the multi-agent LLM pattern to the practical job of turning a paper into a conference poster. It splits the work across four agents: one set for pulling and organizing content, one for layout, one for styling with colors and fonts, and a renderer to put it all together. The system also adds a VLM-based rubric to score layout balance, readability, and aesthetic coherence. That specific mix of agents and rubric is the main new element here, since earlier poster tools existed but did not structure the workflow this way or lean on VLMs for the quality check. The agent roles are described clearly enough that someone could reimplement the pipeline without much guesswork, and the focus on mirroring how human designers actually work is a reasonable choice for an applied task. The paper does a straightforward job of explaining why each agent exists and how they hand off work. The evaluation idea using a VLM rubric is also sensible on paper because running full human studies for every new design tool is costly. The soft spots are right where the stress-test note flags them. The abstract states that the system matches content fidelity and beats prior methods on visual design while producing posters that need little human work, yet it gives no quantitative scores, no list of baselines, no dataset size, and no error analysis. More critically, nothing indicates that the VLM rubric was validated against human raters or expert designers, so any claim of superiority rests on an untested proxy. If the VLM simply prefers outputs from similar models, the loop is real and the results become hard to interpret. This work is aimed at researchers who build or use AI tools for academic workflow tasks like conference prep. A reader looking for a ready-to-try system in that narrow area could get value from the agent breakdown and might even test the rubric themselves. Someone hunting for advances in core LLM reasoning or multi-agent theory will not find much here. It deserves a serious referee because the task is common, the approach is modular and described plainly, and the evaluation gap is fixable rather than fatal. I would send it for review but ask the authors to add concrete metrics, baseline details, and at least a small human correlation study for the rubric before acceptance.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes PosterGen, a multi-agent LLM framework for generating academic posters from research papers. It consists of Parser and Curator agents to extract and organize content into a storyboard, a Layout agent to map content spatially, Stylist agents to apply design elements such as color and typography, and a Renderer to compose the final output. The authors introduce a VLM-based rubric to measure layout balance, readability, and aesthetic coherence. Experimental results are reported to show that PosterGen matches existing methods in content fidelity while significantly outperforming them in visual design, yielding presentation-ready posters that require only minimal human refinements.

Significance. If the evaluation claims hold after proper validation, the work could offer practical value by reducing the manual effort researchers expend on conference poster preparation. The multi-agent decomposition that explicitly mirrors professional designer workflows is a reasonable extension of LLM capabilities to multi-step creative tasks. The VLM rubric itself, if shown to correlate with human judgments, could serve as a reusable tool for automated aesthetic assessment in related generation problems.

major comments (2)

[Abstract] Abstract: the headline claim that PosterGen 'significantly outperforms existing methods in visual designs' and produces 'presentation-ready' posters rests solely on scores from the newly introduced VLM-based rubric; no quantitative metrics, baseline implementations, dataset statistics, or error analysis are supplied, rendering the superiority assertion impossible to evaluate.
[Evaluation] Evaluation (implied by abstract description of VLM rubric): the rubric for layout balance, readability, and aesthetic coherence is presented without any reported human validation, inter-rater reliability statistics, or correlation analysis against expert designer ratings; because this rubric is the only quantitative bridge to the 'minimal human refinements' conclusion, absence of such grounding directly weakens the central experimental claim.

minor comments (1)

[Abstract] Abstract: the collaboration protocol among the four agent types (e.g., message passing or iterative refinement steps) is described at a high level only; a short diagram or pseudocode would clarify the workflow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and describe the revisions we will make to improve the clarity and rigor of our evaluation.

read point-by-point responses

Referee: [Abstract] Abstract: the headline claim that PosterGen 'significantly outperforms existing methods in visual designs' and produces 'presentation-ready' posters rests solely on scores from the newly introduced VLM-based rubric; no quantitative metrics, baseline implementations, dataset statistics, or error analysis are supplied, rendering the superiority assertion impossible to evaluate.

Authors: The abstract is intentionally concise and summarizes the primary findings from the full experimental section. The manuscript body details the VLM rubric scoring procedure, direct comparisons against baseline methods on the same rubric dimensions, and qualitative results demonstrating reduced need for refinements. We agree that the abstract could better signal the evaluation approach. In the revision we will update the abstract to briefly reference the VLM-based quantitative comparison and will add explicit dataset statistics plus a short error analysis subsection in the experiments. revision: yes
Referee: [Evaluation] Evaluation (implied by abstract description of VLM rubric): the rubric for layout balance, readability, and aesthetic coherence is presented without any reported human validation, inter-rater reliability statistics, or correlation analysis against expert designer ratings; because this rubric is the only quantitative bridge to the 'minimal human refinements' conclusion, absence of such grounding directly weakens the central experimental claim.

Authors: We acknowledge that human validation of the VLM rubric would provide stronger support for the automated scores and the 'minimal human refinements' claim. The current manuscript presents the rubric as a scalable proxy aligned with design principles. To address the concern directly, the revised version will include a new human study with expert designers: we will report inter-rater reliability statistics and Pearson/Spearman correlations between VLM rubric scores and human ratings, thereby grounding the evaluation. revision: yes

Circularity Check

0 steps flagged

No circularity: system description with independent empirical evaluation

full rationale

The paper proposes a multi-agent LLM architecture (Parser, Curator, Layout, Stylist, Renderer agents) for paper-to-poster generation and introduces a separate VLM-based rubric to score layout balance, readability, and aesthetic coherence. No equations, derivations, fitted parameters, or predictions appear in the abstract or described workflow. The experimental claim of outperformance is an empirical comparison against baselines using the new rubric; the rubric itself is not shown to be defined in terms of the generated outputs or to reduce to the same inputs by construction. No self-citations are load-bearing for any uniqueness theorem or ansatz. The derivation chain is therefore self-contained as an engineering proposal plus external validation.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework assumes LLMs can reliably perform design tasks when decomposed into agents and that a separate VLM can objectively score aesthetic quality; no free parameters or invented entities are explicitly stated in the abstract.

axioms (2)

domain assumption Multi-agent decomposition of design tasks improves output quality over single-model generation
Invoked in the description of the four collaborative agents mirroring professional workflows.
domain assumption VLM-based rubric provides reliable proxy for human aesthetic judgment
Used to measure layout balance, readability, and aesthetic coherence without further validation mentioned.

pith-pipeline@v0.9.0 · 5760 in / 1204 out tokens · 24014 ms · 2026-05-18T21:59:41.276055+00:00 · methodology

discussion (0)

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Narrative-Driven Paper-to-Slide Generation via ArcDeck
cs.AI 2026-04 unverdicted novelty 6.0

ArcDeck models paper-to-slide generation as narrative reconstruction using discourse parsing and multi-agent refinement, plus a new ArcBench benchmark, to improve flow and coherence over direct summarization.
Quantifying Trust: Financial Risk Management for Trustworthy AI Agents
cs.AI 2026-04 unverdicted novelty 6.0

The paper introduces the Agentic Risk Standard (ARS) as a payment settlement framework that delivers predefined compensation for AI agent execution failures, misalignment, or unintended outcomes.
SciPostGen: Bridging the Gap between Scientific Papers and Poster Layouts
cs.CV 2025-11 conditional novelty 6.0

SciPostGen supplies a paired dataset linking paper structure to poster layouts and shows that retrieval of matching layouts improves generation while respecting user constraints.
AI for Auto-Research: Roadmap & User Guide
cs.AI 2026-05 unverdicted novelty 4.0

The paper delivers a stage-by-stage roadmap for AI in research, showing reliable assistance in retrieval and tool tasks but fragility in novelty and judgment, advocating human-governed collaboration.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · cited by 4 Pith papers

[1]

Title: The main title of the paper

work page
[2]

A Study of Machine Learning Methods

Authors: All author names using initials (no affiliations, emails, or other metadata) Strict Formatting Requirements: •Title: Use proper title case where each word has only the first letter capitalized, EXCEPT for established acronyms, technical terms, or proper nouns that are conventionally written in all uppercase letters (such as abbreviations for orga...

work page
[3]

key visual:Most important method visual representing core research innovation (max 1, middle column)

work page
[4]

problem illustration: Visuals showing research problem, challenges, or motivation (left column introduction)

work page
[5]

method workflow:Method architecture, system diagrams, algorithmic workflows (middle column method)

work page
[6]

main results: Primary experimental results, performance tables, key findings (right column)

work page
[7]

comparative results:Baseline comparisons, ablation studies, validation charts (right column)

work page
[8]

what’s wrong

supporting: Background concepts, supplementary analysis, minor details (flexible placement) Classification Guidelines: •Problem Context: Figures showing “what’s wrong" or “why this matters"Ñ problem illustration •Method Core: Most important technical diagramÑ key visual •Method Details: Architecture/workflow diagramsÑ method workflow •Primary Evidence: Ma...

work page
[9]

Identify Major Sections: •Introduction/Background •Related Work (if substantial) •Methodology/Approach •Experiments/Results •Discussion/Analysis

work page
[10]

Content Processing: •Extract the main content for each section •Keep section content under1000 words •Preserve key technical details, formulas, and findings •Maintain important bullet points and lists •Remove excessive citations and references

work page
[11]

p a p e r _ s e c t i o n s

Section Classification: •foundation: Introduction, background, motivation, problem statement •method: Methodology, approach, algorithm, system design •evaluation: Experiments, results, analysis, validation Required JSON structure: 1 { 2 " p a p e r _ s e c t i o n s " : [ 3 { 4 " section_name " : " I n t r o d u c t i o n " , 5 " s e c t i o n _ t y p e "...

work page
[12]

Left Column Strategy - Foundation & Context: •Introduction/Background/Motivation (priority placement) •Problem definition and challenges •Related work and background context •Method overview or workflow diagrams

work page
[13]

Middle Column Strategy - Core Technical Content: •Primary methodology (highest priority content) •Technical details and algorithms •Theoretical analysis and key innovations •System architecture diagrams

work page
[14]

few-shot example strategy

Right Column Strategy - Experiments & Results: •Experimental results (tables and performance charts) •Key findings and validation data •Performance comparisons and analysis Figure 12 Part 1 of the Curator Agent prompt, focusing on the mission, inputs, and high-level human design patterns. few-shot example strategy. For high scores (4 and 5), we provide po...

work page 2025
[15]

key visual

Key Visual Mandatory Placement: •Identify the “key visual” from classified visuals. This is the MOST important visual •Place key visual in middle column, top priority section •This anchors the entire poster layout around the core research contribution

work page
[16]

Column-Based Visual Distribution: •Column 1 (Left) - Foundation & Context: – MINIMUM:1 visual asset required – Purpose:Express core research problem or contradiction visually –Selection Priority:Choose visuals that illustrate problem context, background concepts, or prior work limitations – Maximum:2 visual assets •Column 2 (Middle) - Methodology: – MANDA...

work page
[17]

Visual Distribution Enforcement:

work page
[18]

Core Task:Create 5-8 poster sections with BOTH content organization AND strategic spatial placement to achieve perfect space utilization across all three columns

Column Space Optimization Strategy:... Core Task:Create 5-8 poster sections with BOTH content organization AND strategic spatial placement to achieve perfect space utilization across all three columns. DO NOT create any conclusion, takeaway, future work, or impact sections. Focus ONLY on problem, method, and results/experiments. Content Organization Guidelines:

work page
[19]

Our Method

Section Requirements: •Section titles: Maximum 4 words (e.g., “Our Method”, “Key Results”) •Text content: 2-3 concise entries using different rich hierarchical formatting (see examples below) based on section contents •Visual integration: Each visual assigned to exactly ONE section •Complete content: No ellipsis (...), write full bullet points

work page
[20]

* Primary c o n c e p t o r f i n d i n g

Rich Text Formatting Options: •A) Nested Bullet Structure: 1 "* Primary c o n c e p t o r f i n d i n g " , 2 " - S u p p o r t i n g d e t a i l o r sub - p o i n t " , 3 " - A d d i t i o n a l s u p p o r t i n g e v i d e n c e " •Other formats like Bold Headers and Ordered Lists are also available. Figure 13 Part 2 of the Curator Agent prompt, specif...

work page
[21]

Left Column: Foundation & Context •Purpose: Introduction, background, prior work, problem setup, supporting context •Content Types: Motivation, challenges, related work, problem definitions, supporting materials •Reading Role: Sets up the research problem and provides necessary background

work page
[22]

NEVER remove method sections

Middle Column: Core Methodology •Purpose: Method details, algorithms, implementation, technical innovation •Content Types:Core methods, algorithms, technical approach, key innovations •Reading Role: Presents the technical contribution and methodology •CRITICAL: Contains key visual (importance level=1). NEVER remove method sections

work page
[23]

Right Column: Results & Impact •Purpose: Experiments, evaluation, findings, conclusions, future work •Content Types:Experimental results, performance analysis, conclusions, future directions •Reading Role: Demonstrates validation and impact of the proposed method Figure 15 Part 1 of the Balancer sub-agent prompt, outlining its role, the current column sta...

work page
[24]

Strategy A: Conservative Text Content Adjustment (for 80-100% utilization) •When to use:Column utilization is close to optimal range (80-100%) •Actions allowed: –MINIMAL text expansion: Add only 1-2 short phrases to underutilized columns (75-85%) –Aggressive text reduction: Significantly shorten content in overflow columns (ą95%) –CONSERVATIVE APPROACH: P...

work page
[25]

Strategy B: Section Management (foră80% or ą100% utilization) •When to use:Column has severe underutilization (ă80%) or overflow (ą100%) •Actions allowed: –Add sections from structured sections: Use additional content from paper sections that fit the column’s purpose –Remove less important sections: Remove sections withimportance level “ 3 or lower import...

work page
[26]

NO CROSS-COLUMN MOVES:Never change column assignment for any existing section

work page
[27]

PRESERVE READING FLOW:Maintain left→middle→right logical progression

work page
[28]

SECTION ID PRESERVATION:Never change section id, section title, visual assets, or other identifying fields

work page
[29]

IMPORTANCE RESPECT:Never remove critical sections (importance level=1 or core results)

work page
[30]

t e x t _ c o n t e n t

TARGET UTILIZATION:Achieve 85-95% utilization for each column Input: {{structured sections}, {current story board}, {column analysis}} Output Format: Output the complete optimized story board JSON. Each section’s ‘text content’ must be an array of complete strings only: 1 " t e x t _ c o n t e n t " : [ 2 "* ** P o i n t T i t l e : * * Complete d e s c r...

work page
[31]

Primary Color Identification: •Look for the main brand color of the organization •Ignore pure white, black, and very light grays (background/outline colors) •Focus on colored elements that define the logo’s visual identity •Consider text colors, graphic elements, symbols, and emblematic elements

work page
[32]

Color Suitability Assessment: •Too Bright: If the main color is very bright/saturated (e.g., neon yellow #FFFF00), generate a more subdued version •Appropriate Saturation: Aim for colors that are vibrant but professional •Readability: Ensure the color provides sufficient contrast on white backgrounds for text

work page
[33]

e x t r a c t e d _ c o l o r

Color Adjustment Rules: •If original color is too bright (lightnessą 85% or saturationą 90%), reduce brightness by 15-25% •If original color is too dark (lightnessą 25%), lighten slightly for better visibility •Maintain the color’s hue character while optimizing for poster applications Output Requirements:Return ONLY a JSON object with the following struc...

work page
[34]

BOLD + CONTRAST COLOR: •Purpose: Core method/methodology names that represent the paper’s unique contribution •Criteria: Novel algorithms, architectures, or techniques introduced by this work; the main methodological innovation that defines the paper; must be unique to this research (not generic terms) •Limit: Maximum 2 per section, prefer 1 if it capture...

work page
[35]

95% accuracy

BOLD: •Purpose: Important quantitative results and core technical terms within each section •Criteria: Performance metrics and numerical results (e.g., “95% accuracy", “5.2ˆ speedup"); key technical concepts central to understanding the section; architecture names, dataset names, established method names; word-level emphasis, not entire phrases •Limit: Ma...

work page
[36]

This was theonly experiment

ITALIC: •Purpose: Defining terms, single-word emphasis, and foreign terminology •Criteria: Technical terms being defined or introduced for the first time; single-word emphasis (e.g., “This was theonly experiment"); foreign words, Latin terms, or specialized vocabulary; word-level application only, never entire sentences •Limit: Maximum 2 per section Outpu...

work page 2022

[1] [1]

Title: The main title of the paper

work page

[2] [2]

A Study of Machine Learning Methods

Authors: All author names using initials (no affiliations, emails, or other metadata) Strict Formatting Requirements: •Title: Use proper title case where each word has only the first letter capitalized, EXCEPT for established acronyms, technical terms, or proper nouns that are conventionally written in all uppercase letters (such as abbreviations for orga...

work page

[3] [3]

key visual:Most important method visual representing core research innovation (max 1, middle column)

work page

[4] [4]

problem illustration: Visuals showing research problem, challenges, or motivation (left column introduction)

work page

[5] [5]

method workflow:Method architecture, system diagrams, algorithmic workflows (middle column method)

work page

[6] [6]

main results: Primary experimental results, performance tables, key findings (right column)

work page

[7] [7]

comparative results:Baseline comparisons, ablation studies, validation charts (right column)

work page

[8] [8]

what’s wrong

supporting: Background concepts, supplementary analysis, minor details (flexible placement) Classification Guidelines: •Problem Context: Figures showing “what’s wrong" or “why this matters"Ñ problem illustration •Method Core: Most important technical diagramÑ key visual •Method Details: Architecture/workflow diagramsÑ method workflow •Primary Evidence: Ma...

work page

[9] [9]

Identify Major Sections: •Introduction/Background •Related Work (if substantial) •Methodology/Approach •Experiments/Results •Discussion/Analysis

work page

[10] [10]

Content Processing: •Extract the main content for each section •Keep section content under1000 words •Preserve key technical details, formulas, and findings •Maintain important bullet points and lists •Remove excessive citations and references

work page

[11] [11]

p a p e r _ s e c t i o n s

Section Classification: •foundation: Introduction, background, motivation, problem statement •method: Methodology, approach, algorithm, system design •evaluation: Experiments, results, analysis, validation Required JSON structure: 1 { 2 " p a p e r _ s e c t i o n s " : [ 3 { 4 " section_name " : " I n t r o d u c t i o n " , 5 " s e c t i o n _ t y p e "...

work page

[12] [12]

Left Column Strategy - Foundation & Context: •Introduction/Background/Motivation (priority placement) •Problem definition and challenges •Related work and background context •Method overview or workflow diagrams

work page

[13] [13]

Middle Column Strategy - Core Technical Content: •Primary methodology (highest priority content) •Technical details and algorithms •Theoretical analysis and key innovations •System architecture diagrams

work page

[14] [14]

few-shot example strategy

Right Column Strategy - Experiments & Results: •Experimental results (tables and performance charts) •Key findings and validation data •Performance comparisons and analysis Figure 12 Part 1 of the Curator Agent prompt, focusing on the mission, inputs, and high-level human design patterns. few-shot example strategy. For high scores (4 and 5), we provide po...

work page 2025

[15] [15]

key visual

Key Visual Mandatory Placement: •Identify the “key visual” from classified visuals. This is the MOST important visual •Place key visual in middle column, top priority section •This anchors the entire poster layout around the core research contribution

work page

[16] [16]

Column-Based Visual Distribution: •Column 1 (Left) - Foundation & Context: – MINIMUM:1 visual asset required – Purpose:Express core research problem or contradiction visually –Selection Priority:Choose visuals that illustrate problem context, background concepts, or prior work limitations – Maximum:2 visual assets •Column 2 (Middle) - Methodology: – MANDA...

work page

[17] [17]

Visual Distribution Enforcement:

work page

[18] [18]

Core Task:Create 5-8 poster sections with BOTH content organization AND strategic spatial placement to achieve perfect space utilization across all three columns

Column Space Optimization Strategy:... Core Task:Create 5-8 poster sections with BOTH content organization AND strategic spatial placement to achieve perfect space utilization across all three columns. DO NOT create any conclusion, takeaway, future work, or impact sections. Focus ONLY on problem, method, and results/experiments. Content Organization Guidelines:

work page

[19] [19]

Our Method

Section Requirements: •Section titles: Maximum 4 words (e.g., “Our Method”, “Key Results”) •Text content: 2-3 concise entries using different rich hierarchical formatting (see examples below) based on section contents •Visual integration: Each visual assigned to exactly ONE section •Complete content: No ellipsis (...), write full bullet points

work page

[20] [20]

* Primary c o n c e p t o r f i n d i n g

Rich Text Formatting Options: •A) Nested Bullet Structure: 1 "* Primary c o n c e p t o r f i n d i n g " , 2 " - S u p p o r t i n g d e t a i l o r sub - p o i n t " , 3 " - A d d i t i o n a l s u p p o r t i n g e v i d e n c e " •Other formats like Bold Headers and Ordered Lists are also available. Figure 13 Part 2 of the Curator Agent prompt, specif...

work page

[21] [21]

Left Column: Foundation & Context •Purpose: Introduction, background, prior work, problem setup, supporting context •Content Types: Motivation, challenges, related work, problem definitions, supporting materials •Reading Role: Sets up the research problem and provides necessary background

work page

[22] [22]

NEVER remove method sections

Middle Column: Core Methodology •Purpose: Method details, algorithms, implementation, technical innovation •Content Types:Core methods, algorithms, technical approach, key innovations •Reading Role: Presents the technical contribution and methodology •CRITICAL: Contains key visual (importance level=1). NEVER remove method sections

work page

[23] [23]

Right Column: Results & Impact •Purpose: Experiments, evaluation, findings, conclusions, future work •Content Types:Experimental results, performance analysis, conclusions, future directions •Reading Role: Demonstrates validation and impact of the proposed method Figure 15 Part 1 of the Balancer sub-agent prompt, outlining its role, the current column sta...

work page

[24] [24]

Strategy A: Conservative Text Content Adjustment (for 80-100% utilization) •When to use:Column utilization is close to optimal range (80-100%) •Actions allowed: –MINIMAL text expansion: Add only 1-2 short phrases to underutilized columns (75-85%) –Aggressive text reduction: Significantly shorten content in overflow columns (ą95%) –CONSERVATIVE APPROACH: P...

work page

[25] [25]

Strategy B: Section Management (foră80% or ą100% utilization) •When to use:Column has severe underutilization (ă80%) or overflow (ą100%) •Actions allowed: –Add sections from structured sections: Use additional content from paper sections that fit the column’s purpose –Remove less important sections: Remove sections withimportance level “ 3 or lower import...

work page

[26] [26]

NO CROSS-COLUMN MOVES:Never change column assignment for any existing section

work page

[27] [27]

PRESERVE READING FLOW:Maintain left→middle→right logical progression

work page

[28] [28]

SECTION ID PRESERVATION:Never change section id, section title, visual assets, or other identifying fields

work page

[29] [29]

IMPORTANCE RESPECT:Never remove critical sections (importance level=1 or core results)

work page

[30] [30]

t e x t _ c o n t e n t

TARGET UTILIZATION:Achieve 85-95% utilization for each column Input: {{structured sections}, {current story board}, {column analysis}} Output Format: Output the complete optimized story board JSON. Each section’s ‘text content’ must be an array of complete strings only: 1 " t e x t _ c o n t e n t " : [ 2 "* ** P o i n t T i t l e : * * Complete d e s c r...

work page

[31] [31]

Primary Color Identification: •Look for the main brand color of the organization •Ignore pure white, black, and very light grays (background/outline colors) •Focus on colored elements that define the logo’s visual identity •Consider text colors, graphic elements, symbols, and emblematic elements

work page

[32] [32]

Color Suitability Assessment: •Too Bright: If the main color is very bright/saturated (e.g., neon yellow #FFFF00), generate a more subdued version •Appropriate Saturation: Aim for colors that are vibrant but professional •Readability: Ensure the color provides sufficient contrast on white backgrounds for text

work page

[33] [33]

e x t r a c t e d _ c o l o r

Color Adjustment Rules: •If original color is too bright (lightnessą 85% or saturationą 90%), reduce brightness by 15-25% •If original color is too dark (lightnessą 25%), lighten slightly for better visibility •Maintain the color’s hue character while optimizing for poster applications Output Requirements:Return ONLY a JSON object with the following struc...

work page

[34] [34]

BOLD + CONTRAST COLOR: •Purpose: Core method/methodology names that represent the paper’s unique contribution •Criteria: Novel algorithms, architectures, or techniques introduced by this work; the main methodological innovation that defines the paper; must be unique to this research (not generic terms) •Limit: Maximum 2 per section, prefer 1 if it capture...

work page

[35] [35]

95% accuracy

BOLD: •Purpose: Important quantitative results and core technical terms within each section •Criteria: Performance metrics and numerical results (e.g., “95% accuracy", “5.2ˆ speedup"); key technical concepts central to understanding the section; architecture names, dataset names, established method names; word-level emphasis, not entire phrases •Limit: Ma...

work page

[36] [36]

This was theonly experiment

ITALIC: •Purpose: Defining terms, single-word emphasis, and foreign terminology •Criteria: Technical terms being defined or introduced for the first time; single-word emphasis (e.g., “This was theonly experiment"); foreign words, Latin terms, or specialized vocabulary; word-level application only, never entire sentences •Limit: Maximum 2 per section Outpu...

work page 2022