Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models

Hugging Face Daily Papers Papers

Summary

This paper introduces SEIG, a framework that uses pretrained vision-language models to reconstruct 3D scenes from single images as editable Blender programs through progressive refinement of geometry, materials, composition, and lighting.

Inverse graphics is a longstanding and highly underconstrained problem that seeks to reconstruct images as editable 3D scenes which can be rendered, relit, and manipulated. In this work, we investigate whether pretrained vision-language models (VLMs) can perform executable inverse graphics directly from a single image by reconstructing a scene as an editable Blender program, without relying on specialized 2D or 3D foundation models, differentiable rendering, or multi-view supervision. We introduce Staged Executable Inverse Graphics (SEIG), an agentic framework that reconstructs a 3D scene from a single image by progressively refining scene factors including geometry, materials, composition, and lighting directly in executable Blender code space. We evaluate our framework across diverse scenes using a range of reconstruction metrics spanning pixel-level, perceptual, and semantic fidelity. Our experiments show that staged reconstruction substantially improves reconstruction fidelity, highlighting the importance of task decomposition for executable inverse graphics with general-purpose VLMs. Finally, we showcase various downstream applications enabled by the reconstructed editable Blender scenes.
Original Article
View Cached Full Text

Cached at: 06/02/26, 03:37 PM

Paper page - Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models

Source: https://huggingface.co/papers/2606.02580

Abstract

Pretrained vision-language models can reconstruct 3D scenes from single images as editable Blender programs through progressive refinement, demonstrating improved fidelity through staged reconstruction approaches.

Inverse graphicsis a longstanding and highly underconstrained problem that seeks to reconstruct images as editable 3D scenes which can be rendered, relit, and manipulated. In this work, we investigate whether pretrainedvision-language models(VLMs) can performexecutable inverse graphicsdirectly from a single image by reconstructing a scene as an editableBlender program, without relying on specialized 2D or 3D foundation models, differentiable rendering, or multi-view supervision. We introduce StagedExecutable Inverse Graphics(SEIG), an agentic framework that reconstructs a 3D scene from a single image by progressively refiningscene factorsincludinggeometry,materials,composition, andlightingdirectly in executable Blender code space. We evaluate our framework across diverse scenes using a range of reconstruction metrics spanning pixel-level, perceptual, and semantic fidelity. Our experiments show thatstaged reconstructionsubstantially improves reconstruction fidelity, highlightingthe importance oftask decompositionforexecutable inverse graphicswith general-purpose VLMs. Finally, we showcase various downstream applications enabled by the reconstructed editable Blender scenes.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2606\.02580

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.02580 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.02580 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.02580 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

InSight: Self-Guided Skill Acquisition via Steerable VLAs

Hugging Face Daily Papers

InSight presents a framework for autonomous skill acquisition in vision-language-action (VLA) models by enabling steerability at the primitive-action level and using a VLM-guided data flywheel to generate demonstrations, achieving manipulation tasks like block flipping and pouring without human demonstrations.

Stateful Visual Encoders for Vision-Language Models

Hugging Face Daily Papers

This paper introduces a stateful visual encoder for vision-language models that conditions visual representations on prior features, enabling better visual comparison in multi-image and agentic settings. The method shows consistent improvements across tasks such as cross-image spatial aggregation and longitudinal radiology.