Organizations: School of Cybersecurity, Northwestern Polytechnical University · Shanghai Artificial Intelligence Laboratory · Shanghai Innovation Institution · Fudan University
Producing presentation slides automatically entails coordinating narrative structure with page-level graphic design under strict spatial constraints. For such structured multimodal tasks, a well-organized design process is essential to ensure the final quality of slides. Existing approaches rely on fixed templates or directly emit executable code, thereby both limiting the creative layout-design capabilities of LLMs and bypassing the essential slide-page design step. To address these limitations, this paper (1) proposes a hierarchical slides generation workflow, DeepSlides, that systematically organizes slide design tasks without any predefined template or style, decoupling slide-page design from implementation; (2) introduces SlideDesign, a dataset tailored specifically for slides generation tasks; and (3) presents a multi-agent reinforcement learning training paradigm and trains a couple of models, SlideQwens, for slide design and implementation. Experimental results demonstrate that our proposed framework outperforms baseline methods on evaluated metrics and achieves superior performance in human preference evaluations. The dataset and code are available at https://github.com/sxswz213/DeepSlides.
Figures & tables
Figure 1: Overview of the DeepSlides workflow. Given a topic and optional user-specified page-level requirements, the system first determines the high-level slide style. Subsequently, each slide is generated through iterative refinement.
Figure 2: Comparative visualization of slide decks generated for the topic "Urban Heat Island Mitigation", contrasting slides produced by direct model generation, Auto-Slides, EvoPresent, DeepSlides (ours), and two commercial solutions, Manus and Kimi. Additional cases can be found in Appendix F .
Output Format
Complete Presentation
Template Free
EvoPresent
HTML
✓
×
AutoSlides
LaTeX/PDF
✓
×
SlideGen
PPTX
✓
×
SlideCoder
PPTX
×
×
DOC2PPT
PPTX/PDF
✓
×
PPTAgent
PPTX/HTML
✓
×
Table 1: Comparison of slide-generation workflows. light blue rows indicate open-source methods, while light orange rows denote closed-source systems.
Figure 3: Illustration of the DeepSlides hierarchical slides generation workflow. The framework begins by generating a detailed outline through multimodal deep research or directly from user-provided outlines. Subsequently, slide-level style guidelines are defined, including colors, shapes, and backgrounds. At the page-level, the system performs content expansion via web search and executes detailed design across background, layout, and content layers. Python code is generated to implement each slide, followed by iterative refinement based on evaluation and feedback. Finally, completed slides are assembled into a cohesive presentation.
Figure 4: Examples of data in SlideDesign.
Figure 5: Data generation pipeline
Figure 6: Subject taxonomy and dataset composition. Inner ring: 6 primary areas with counts; outer ring: 42 secondary disciplines.
Figure 7: RL train
Method
Objective Metrics
Content
Visual
Success Rate (%)
Balance ( ↑ )
Clarity
Coherence
Layout
Hierarchy
Color
Avg.
Direct ClaudeHaiku
99
0.53
3.99
3.64
2.64
3.87
3.14
3.46
Direct ClaudeSonnet
97
0.67
3.79
3.47
2.67
3.64
3.13
3.34
Auto-Slides ClaudeHaiku
87
0.81
3.53
3.64
2.49
3.50
3.08
3.25
EvoPresent ClaudeHaiku
100
0.77
3.53
3.84
3.46
3.95
3.69
3.69
DeepSlides ClaudeHaiku
95
0.78
3.79
3.81
3.55
4.04
3.71
3.78
Table 2: We compare baseline systems and DeepSlides variants using objective metrics and VLM-as-judge scores for content quality and visual design. Best and second-best results are highlighted in bold and underline , respectively.
Figure 8: We report the fraction of topics for which each framework is ranked #1 by human evaluators.
Figure 9: Pairwise win rate.
Figure 10: Human evaluation scores.
Model
Success Rate
Balance
Visual
Qwen3-0.6B
0.0
-
-
SlideQwens sft
81.4
0.77
3.23
SlideQwens full
86.0
0.77
3.38
Table 3: Model Training Performance Comparison.
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 11: Detailed information on the human evaluation questionnaire.
Figure 12: Detailed information on the human evaluation questionnaire.
Large language models (LLMs) have demonstrated strong potential in agentic tasks, particularly in slide generation. However, slide generation poses a fundamental challenge: the generation process is text-centric, whereas its quality is governed by visual aesthetics. This modality gap leads current models to frequently produce slides with aesthetically suboptimal layouts. Existing solutions typically rely either on heavy visual reflection, which incurs high inference cost yet yields limited gains; or on fine-tuning with large-scale datasets, which still provides weak and indirect aesthetic supervision. In contrast, the explicit use of aesthetic principles as supervision remains unexplored. In this work, we present AeSlides, a reinforcement learning framework with verifiable rewards for Aesthetic layout supervision in Slide generation. We introduce a suite of meticulously designed verifiable metrics to quantify slide layout quality, capturing key layout issues in an accurate, efficient, and low-cost manner. Leveraging these verifiable metrics, we develop a GRPO-based reinforcement learning method that directly optimizes slide generation models for aesthetically coherent layouts. With only 5K training prompts on GLM-4.7-Flash, AeSlides improves aspect ratio compliance from 36% to 85%, while reducing whitespace by 44%, element collisions by 43%, and visual imbalance by 28%. Human evaluation further shows a substantial improvement in overall quality, increasing scores from 3.31 to 3.56 (+7.6%), outperforming both model-based reward optimization and reflection-based agentic approaches, and even edging out Claude-Sonnet-4.5. These results demonstrate that such a verifiable aesthetic paradigm provides an efficient and scalable approach to aligning slide generation with human aesthetic preferences. Our repository is available at https://github.com/ympan0508/aeslides.
Presentations are a primary medium for scholarly communication, yet most AI slide generators optimize the artifact (a visually plausible deck) while under-optimizing the delivery process (pacing, narrative, and presentation preparation). We present DeepSlide, a human-in-the-loop multi-agent system that supports preparing the full presentation process, from requirement elicitation and time-budgeted narrative planning, to evidence-grounded slide--script generation, attention augmentation, and rehearsal support. DeepSlide integrates (i) a controllable logical-chain planner with per-node time budgets, (ii) a lightweight content-tree retriever for grounding, (iii) Markov-style sequential rendering with style inheritance, and (iv) sandboxed execution with minimal repair to ensure renderability. We further introduce a dual-scoreboard benchmark that cleanly separates static artifact quality from dynamic delivery excellence. Across 20 domains and diverse audience profiles, DeepSlide matches strong baselines on artifact quality while consistently achieving larger gains on delivery metrics, improving narrative flow, pacing precision, and slide--script synergy with clearer attention guidance.
Ming Yang, Zhiwei Zhang, Jiahang Li +3
School of Data Science, Fudan University, Shanghai, China
Slide design requires personalizing both deck themes and page layouts. Yet, current AI agent-based methods struggle with fine-grained, page-level design. Solely relying on prespecified templates or user verbose instructions, they fail to capture latent design intents, leaving Page-level Slide Personalization (PSP) unresolved. To close this gap, this work formulates PSP as an inverse planning problem. We propose to learn a design intent without assuming any knowledge of the specific executing tools (e.g., PowerPoint, Beamer) being used. However, relinquishing control over these tools makes the problem intractable to optimize end-to-end. To overcome this, we propose SPIRE, a principled framework to solve PSP approximately. By intentionally corrupting the visual structures of clean slides, SPIRE creates a verifiable task to denoise the corruption, whereby two agents learn to collaboratively refine executable designs via reinforcement learning (RL). We present a proof that structural denoising is a consistent surrogate for PSP, and that the multi-agent formulation strictly reduces policy gradient variance in RL. Extensive experiments demonstrate the superiority of SPIRE.
Tianci Liu, Zihan Dong, Linjun Zhang +5
Purdue University · Microsoft · Rutgers University +1