Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 8 of the 8 references it cites were found in a registry or an index.

From Pixel to Mesh: Accelerating Game Asset Creation via a Semantically-Guided 2D-to-3D Generative Pipeline

Open access

Abstract

In independent game development, creating high-quality 3D assets remains a critical bottleneck, as traditional workflows require specialized skills that hinder non-expert designers. While Generative AI has democratized 2D concept art, translating these visions into game-ready 3D assets is technically demanding. To address this, we propose an automated pipeline that leverages semantic and geometric feature extraction to synthesize 3D meshes and PBR materials from AI-generated 2D images. Our system orchestrates a workflow bridging text-to-image diffusion models with advanced computer vision modules. It employs monocular depth estimation and intrinsic decomposition to interpret geometric structures and isolate material properties (albedo, roughness, metallic) from multi-view consistent concepts. These features are algorithmically processed to produce finalized .glb assets. A comparative user study with indie developers demonstrates that this pipeline reduces asset production time by approximately 80% compared to traditional modeling tools. By shifting the user’s role from manual vertex manipulation to high-level semantic curation, this research validates a novel workflow that empowers designers to rapidly populate immersive worlds, streamlining the future of interactive media design.

Overview

CoChat Summary

TL;DR

The paper proposes a semantically-guided 2D-to-3D generative pipeline that turns text/image concepts into game-ready meshes with PBR materials via multi-view diffusion, depth/normal estimation, intrinsic decomposition, SDF reconstruction, and auto-retopology. A user study with 18 indie developers found large production-time reductions and high usability, while noting limits for complex rigged characters and hero assets.

Key findings

  • In the user study, expert modelers’ average production time dropped from 145 minutes under the traditional workflow to 22 minutes with the proposed pipeline, an efficiency gain of approximately 85%; the abstract/conclusion summarize this as roughly 80% reduction.
  • Task completion rates were 55.5% overall for the traditional condition versus 100% for the proposed pipeline; novices completed only 11% (1/9) traditionally but 100% (9/9) with the pipeline, and all novices generated a usable asset within 30 minutes under Condition B.
  • Blind review by three senior art directors found Condition A scored slightly higher on “Geometry Cleanliness” for experts (M = 6.2 vs. M = 5.8), while Condition B achieved comparable “Texture Quality” (M = 5.9) and significantly higher scores for novices across all metrics.
  • Usability was rated “Excellent” with a mean SUS score of 82.5; NASA-TLX showed reduced physical demand and frustration, 78% of participants praised semantic guidance, and 22% of experts wanted granular vertex-level control.

Methods

  • The system is a three-module pipeline: multi-view consistent 2D concept generation, geometric and material feature extraction (monocular depth, surface normals, intrinsic decomposition into albedo/roughness/metallic), and volumetric reconstruction with Neural SDF, Marching Cubes, auto-retopology, UV unwrapping, and texture baking to .glb/.fbx.
  • Evaluation was a within-subjects comparative study with N = 18 participants stratified into Novice Designers (n = 9, <1 year experience) and Expert Modelers (n = 9, >5 years industry experience).
  • Participants created a “Stylized Sci-Fi Crate” for a Unity-based game under two counterbalanced conditions: Condition A used Blender and Substance Painter; Condition B used the proposed semantically-guided system. Measures included Time-on-Task, task completion, blind fidelity ratings on a 7-point Likert scale, SUS, NASA-TLX, and semi-structured interviews.

Limitations

  • The paper states the pipeline currently has limitations for complex, rigged characters where precise anatomical topology is critical, and that “hero assets” often require manual refinement for deformation quality during animation.
  • The paper also states that stochasticity in diffusion models can occasionally introduce geometric hallucinations, requiring regeneration or manual correction.
  • The text makes evident additional constraints: the evaluation uses a small sample (N = 18), a single asset type (“Stylized Sci-Fi Crate”), and reports “statistically significant” time reduction without presenting inferential test statistics.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.54941/ahfe1007640

Cites

0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Overview of the proposed generative pipeline. The system processes a text prompt through four stages: (1) 2D concept generation, (2) Depth estimation, (3) Normal map prediction, and (4) Final volumetric reconstruction overlaid with wireframe topology.
Figure 1 Overview of the proposed generative pipeline. The system processes a text prompt through four stages: (1) 2D concept generation, (2) Depth estimation, (3) Normal map prediction, and (4) Final volumetric reconstruction overlaid with wireframe topology.From the PDF
Figure 2 Visualization of intermediate feature maps extracted during the generation process. (a) The input 2D concept; (b) Estimated depth map utilizing monocular cues; (c) Predicted surface normal map for geometric detailing; (d) The final physicallybased rendered (PBR) 3D asset.
Figure 2 Visualization of intermediate feature maps extracted during the generation process. (a) The input 2D concept; (b) Estimated depth map utilizing monocular cues; (c) Predicted surface normal map for geometric detailing; (d) The final physicallybased rendered (PBR) 3D asset.From the PDF
Figure 3 Comparison of average asset production time between the Traditional Workflow (Condition A) and the Proposed AI Pipeline (Condition B). Note that novice users in the traditional condition frequently reached the 180-minute timeout limit without successfully completing the task.
Figure 3 Comparison of average asset production time between the Traditional Workflow (Condition A) and the Proposed AI Pipeline (Condition B). Note that novice users in the traditional condition frequently reached the 180-minute timeout limit without successfully completing the task.From the PDF
Figure 4 NASA-TLX workload analysis comparison. The proposed pipeline (blue) demonstrates significantly lower mental and physical demand compared to the traditional workflow (gray), validating the shift from manual execution to semantic curation.
Figure 4 NASA-TLX workload analysis comparison. The proposed pipeline (blue) demonstrates significantly lower mental and physical demand compared to the traditional workflow (gray), validating the shift from manual execution to semantic curation.From the PDF

Verification

Claims the paper makes

0 Valid · 0 Not valid · 0 Not checked · 36 Own results and interpretations

16 more claims

Create a free account to see all 36 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 10
  • It may state two separate claims as one. 6
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

The publisher hasn’t deposited this paper’s reference list.

Metadata

Authors
Hu, J., Li, J., Wang, Z., Li, Z., Chan, K.-C., Ma, X., Jiang, Z., & Zhang, Y.
Affiliations
  • Shanghai University of Engineering Science (CN)
  • Goldsmiths University of London (GB)
  • University of Exeter (GB)
  • UNSW Sydney (AU)
  • University of Hong Kong (HK)
  • Hebei University of Science and Technology (CN)
Venue
AHFE International
Published
2026
Volume
216
ISSN
2771-0718 (ISSN-L)
OpenAlex
W7167362966
Cited by
0
References
0
FWCI
0.00
Type
conference paper
Language
en
Open access
Locations
AHFE international (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › 3D Shape Modeling and Analysis
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Computer Graphics and Visualization Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
Keywords
game asset creation, indie game development, semantic segmentation, monocular depth estimation, intrinsic decomposition, text-to-image diffusion models

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: From Pixel to Mesh: Accelerating Game Asset Creation via a Semantically-Guided 2D-to-3D Generative Pipeline. https://app.cochat.ai/papers/10.54941/ahfe1007640
The paper
Hu, J., Li, J., Wang, Z., Li, Z., Chan, K.-C., Ma, X., Jiang, Z., & Zhang, Y. (2026). From Pixel to Mesh: Accelerating Game Asset Creation via a Semantically-Guided 2D-to-3D Generative Pipeline. AHFE International. https://doi.org/10.54941/ahfe1007640

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.