ProcTex: Consistent and Interactive Text-to-texture Synthesis for Part-based Procedural Models

TL;DR

ProcTex addresses the challenge of consistent and interactive text-guided texture synthesis across families of part-based procedural models. It combines template-based texture synthesis with a texture transfer stage driven by dense surface correspondences learned by ProCorrNet, complemented by a retexturing pipeline and part-level UV maps for local edits. The approach delivers high-quality, visually coherent textures with real-time transfer and supports localized appearance edits, outperforming baseline methods in both perceptual fidelity and cross-shape consistency. Extensive experiments on 22 procedural models demonstrate robust texture transfer, adaptability to structural variations, and interactive performance suitable for iterative procedural design workflows.

Abstract

Recent advances in generative modeling have driven significant progress in text-guided texture synthesis. However, current methods focus on synthesizing texture for single static 3D object, and struggle to handle entire families of shapes, such as those produced by procedural programs. Applying existing methods naively to each procedural shape is too slow to support exploring different parameter configurations at interactive rates, and also results in inconsistent textures across the procedural shapes. To this end, we introduce ProcTex, the first text-to-texture system designed for part-based procedural models. ProcTex enables consistent and real-time text-guided texture synthesis for families of shapes, which integrates seamlessly with the interactive design flow of procedural modeling. To ensure consistency, our core approach is to synthesize texture for a template shape from the procedural model, followed by a texture transfer stage to apply the texture to other procedural shapes via solving dense correspondence. To ensure interactiveness, we propose a novel correspondence network and show that dense correspondence can be effectively learned by a neural network for procedural models. We also develop several techniques, including a retexturing pipeline to support structural variation from procedural parameters, and part-level UV texture map generation for local appearance editing. Extensive experiments on a diverse set of procedural models validate ProcTex's ability to produce high-quality, visually consistent textures while supporting interactive applications.

Figures (4)

• Figure 5: Part appearance editing with texture maps generated by ProcTex. Modify the sofa's back support texture. Left: Original texture; Middle: Texture after a hue shift of $+60^\circ$ in HSV color space; Right: Texture embossed with an eye-shaped logo
• Figure 6: Qualitative comparison of Neural Cages and our modified variant on thin procedural models. The baseline Neural Cage often maps interior surfaces onto the exterior, which creates untextured regions (circled in purple). Our modification resolves these errors, ensuring complete texture coverage and higher visual fidelity, even though standard quantitative correspondence metrics like Chamfer distances (CD) remain nearly identical.
• Figure 7: Qualitative comparison of ProcTex against two baselines (InTex and SDS+Lipschitz) across three prompts. ProcTex produces more consistent and detailed textures across procedural variations, while the baselines often yield distorted or incomplete results.
• Figure 8: More qualitative results. Each row corresponds to one procedural model. Text prompts used from top to bottom: (1) “An ivory chair with brown patterns in the backrest”; (2) “A Japanese ceramic pottery”; (3) "A white shell"; (4) "A yellow starfruit"; (5) "A mug with cat patterns"; (6) "A worn school desk". For the chair model, only a subset of parameters are displayed here; please refer to the supplemental video for more details. These results highlight that our method maintains both global style and local detail across wide procedural variations.