OpenNovelty.org

FieryGS: In-the-Wild Fire Synthesis with Physics-Integrated Gaussian Splatting

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.0Download Report PDF
3D Gaussian SplattingPhysics SimulationCombustion SimulationNovel View Synthesis
Abstract:

We consider the problem of synthesizing photorealistic, physically plausible combustion effects in in-the-wild 3D scenes. Traditional CFD and graphics pipelines can produce realistic fire effects but rely on handcrafted geometry, expert-tuned parameters, and labor-intensive workflows, limiting their scalability to the real world. Recent scene modeling advances like 3D Gaussian Splatting (3DGS) enable high-fidelity real-world scene reconstruction, yet lack physical grounding for combustion. To bridge this gap, we propose FieryGS, a physically-based framework that integrates physically-accurate and user-controllable combustion simulation and rendering within the 3DGS pipeline, enabling realistic fire synthesis for real scenes. Our approach tightly couples three key modules: (1) multimodal large-language-model-based physical material reasoning, (2) efficient volumetric combustion simulation, and (3) a unified renderer for fire and 3DGS. By unifying reconstruction, physical reasoning, simulation, and rendering, FieryGS removes manual tuning and automatically generates realistic, controllable fire dynamics consistent with scene geometry and materials. Our framework supports complex combustion phenomena—including flame propagation, smoke dispersion, and surface carbonization—with precise user control over fire intensity, airflow, ignition location and other combustion parameters. Evaluated on diverse indoor and outdoor scenes, FieryGS outperforms all comparative baselines in visual realism, physical fidelity, and controllability.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper proposes FieryGS, a framework integrating physically-based combustion simulation with 3D Gaussian Splatting to synthesize photorealistic fire in reconstructed real-world scenes. It resides in the 'Physics-Integrated Fire Synthesis in 3D Scenes' leaf, which contains only three papers total, including this work. This is a notably sparse research direction within the broader taxonomy of forty papers, suggesting the intersection of physics-driven combustion modeling and modern neural scene reconstruction remains relatively unexplored compared to volumetric reconstruction or standalone simulation branches.

The taxonomy reveals several neighboring research directions. Volumetric Fire Reconstruction from Multi-View Imagery (fifteen papers across three sub-branches) focuses on capturing real flames from camera arrays using tomographic or neural implicit methods, prioritizing fidelity to observed data rather than synthesis in novel scenes. Procedural and Data-Driven Fire Generation (six papers) emphasizes algorithmic or learning-based flame appearance without exhaustive simulation. Combustion Simulation and Modeling (seven papers) addresses detailed numerical solvers for reactive flows but typically lacks integration with scene reconstruction pipelines. FieryGS bridges physics simulation and neural rendering, diverging from pure reconstruction or procedural approaches.

Among fifteen candidates examined across three contributions, none were flagged as clearly refuting the work. The core FieryGS framework examined seven candidates with zero refutable overlaps, as did the unified volumetric renderer contribution. The MLLM-based material reasoning component examined only one candidate, also non-refutable. This limited search scope—fifteen papers from semantic retrieval and citation expansion—suggests the analysis captures nearby work but cannot claim exhaustive coverage. The absence of refutable candidates within this sample indicates the specific combination of multimodal material reasoning, volumetric combustion, and 3DGS integration appears relatively unexplored in the examined literature.

Based on the top-fifteen semantic matches and the sparse three-paper leaf structure, the work appears to occupy a distinct position combining physical simulation with modern neural scene representations. The limited search scope and small sibling set mean this assessment reflects local novelty within examined candidates rather than definitive field-wide uniqueness. Broader searches or domain-specific combustion graphics venues might reveal additional related efforts not captured here.

Taxonomy

Core-task Taxonomy Papers
40
3
Claimed Contributions
15
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: Photorealistic fire synthesis in reconstructed 3D scenes. The field encompasses a diverse set of approaches organized around several main branches. Volumetric Fire Reconstruction from Multi-View Imagery focuses on capturing real flame phenomena through tomographic and multi-view techniques, often leveraging chemiluminescence or radiative measurements to reconstruct 3D flame structures from camera arrays. Physics-Integrated Fire Synthesis in 3D Scenes emphasizes coupling physical combustion models with rendering pipelines to achieve realism grounded in thermodynamic and fluid-dynamic principles. Procedural and Data-Driven Fire Generation explores algorithmic and learning-based methods for generating flame appearances without exhaustive simulation, while Combustion Simulation and Modeling delves into detailed numerical solvers for reactive flows. Flame Geometry Characterization and Diagnostics targets the measurement and analysis of flame shapes and properties, and Application-Specific Fire Visualization and Simulation addresses domain-tailored solutions such as disaster response, virtual reality training, and interactive graphics. Recent work reveals contrasting emphases between reconstruction-driven and synthesis-driven paradigms. Reconstruction methods like Swirling Flame Reconstruction[5] and Flame Tomographic Reconstruction[15] prioritize fidelity to observed data, whereas synthesis approaches such as FieryGS[0] and FlameGS[7] aim to generate plausible fire within novel 3D scenes by integrating physical priors or neural representations. FieryGS[0] sits squarely within Physics-Integrated Fire Synthesis, combining Gaussian splatting with combustion-aware modeling to produce photorealistic flames in reconstructed environments. This contrasts with purely procedural techniques and also differs from purely data-driven reconstruction pipelines that lack explicit physical constraints. Nearby works like Fire AR Transfer[2] and Dynamic Flame Illumination[27] explore interactive augmented reality and illumination effects, highlighting ongoing questions about balancing computational efficiency, physical accuracy, and visual plausibility across diverse application contexts.

Claimed Contributions

FieryGS framework for physically-based fire synthesis in 3DGS

The authors introduce FieryGS, a framework that combines physically-grounded combustion simulation with 3D Gaussian Splatting to generate photorealistic and controllable fire effects in real-world scenes. The framework unifies reconstruction, physical reasoning, simulation, and rendering to automatically produce realistic fire dynamics consistent with scene geometry and materials.

7 retrieved papers
MLLM-based physical material reasoning for combustion properties

The method employs multimodal large language models to perform zero-shot inference of combustion-relevant material properties from 3DGS reconstructions. This enables automatic identification of material types, burnability, thermal diffusivity, and smoke color without manual annotation.

1 retrieved paper
Unified volumetric renderer for fire and 3DGS integration

The authors develop a novel unified volumetric rendering pipeline that seamlessly integrates simulated fire, smoke, and reconstructed 3DGS scenes. This renderer accounts for smoke scattering, fire illumination, and surface carbonization to produce photorealistic combustion effects.

7 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

FieryGS framework for physically-based fire synthesis in 3DGS

The authors introduce FieryGS, a framework that combines physically-grounded combustion simulation with 3D Gaussian Splatting to generate photorealistic and controllable fire effects in real-world scenes. The framework unifies reconstruction, physical reasoning, simulation, and rendering to automatically produce realistic fire dynamics consistent with scene geometry and materials.

Contribution

MLLM-based physical material reasoning for combustion properties

The method employs multimodal large language models to perform zero-shot inference of combustion-relevant material properties from 3DGS reconstructions. This enables automatic identification of material types, burnability, thermal diffusivity, and smoke color without manual annotation.

Contribution

Unified volumetric renderer for fire and 3DGS integration

The authors develop a novel unified volumetric rendering pipeline that seamlessly integrates simulated fire, smoke, and reconstructed 3DGS scenes. This renderer accounts for smoke scattering, fire illumination, and surface carbonization to produce photorealistic combustion effects.

Table of Contents