Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models (Jun 2026)
Title: Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models (Jun 2026) Link: https://arxiv.org/abs/2606.02580 Date: June 1, 2026 Summary: This paper introduces SEIG, an agentic framework that reconstructs editable 3D Blender scenes from a single image using a pretrained vision-language model without specialized 3D tools. By decomposing the reconstruction into sequential stages—geometry, materials, composition, and lighting—and employing a generator-verifier loop, it produces structured code suitable for novel-view synthesis, editing, and physics simulation. Key Topics: - Inverse Graphics - Vision-Language Models - 3D Scene Reconstruction - Agentic Frameworks - Blender Chapters: 00:00 - Introduction & Core Insight 01:46 - Defining Inverse Graphics Problems 03:45 - NeRFs vs Symbolic Code 05:45 - Staged Reconstruction Framework 06:50 - Generator-Verifier Feedback Loop 08:41 - Testing Against Baseline Models 10:22 - Why VLMs Outperform Specialists 11:51 - Reconstructing Complex Visual Scenes 13:10 - Spatial Hierarchy and Composition 14:41 - Granular Editing and Relighting 15:41 - Physics Simulation Capabilities 16:13 - Future Robotics Implications Stock video credits: - Nicola Narracci - https://www.pexels.com/@nicola-narracci-157460431 - Sajjad Sabbir - https://www.pexels.com/@sajjad-sabbir-2151690825 - Chandresh Uike - https://www.pexels.com/@chandresh-uike-754623426 - Sasha Poberailo - https://www.pexels.com/@sasha-poberailo-113008664 - Shutter Break - https://www.pexels.com/@shutterbreak - Кирилл Левченко - https://www.pexels.com/@2156561057 - Computer_Scientist - https://www.pexels.com/@computer_scientist-3594007 - cottonbro studio - https://www.pexels.com/@cottonbro - Rafael Minguet Delgado - https://www.pexels.com/@thales13 - The MoonRunners - https://www.pexels.com/@themoonrunnerrs - Ron Lach - https://www.pexels.com/@ron-lach - Nitin Khajotia - https://www.pexels.com/@nitin-khajotia-55235902 - jorguez - https://www.pexels.com/@jorguez-191283684 - Pressmaster - https://www.pexels.com/@pressmaster - Marina Leonova - https://www.pexels.com/@marina-zasorina - Jakub Zerdzicki - https://www.pexels.com/@jakubzerdzicki - Coverr - https://www.pexels.com/@coverr - Thirdman - https://www.pexels.com/@thirdman - KoolShooters - https://www.pexels.com/@koolshooters - Kindel Media - https://www.pexels.com/@kindelmedia - Joshua Malic - https://www.pexels.com/@joshua-malic-25131152 - Life Of Pix - https://www.pexels.com/@life-of-pix - PNW Production - https://www.pexels.com/@pnw-prod - Vladislav Styazhkin - https://www.pexels.com/@vladislav-styazhkin-157457354 - John Smith - https://www.pexels.com/@john-smith-135446067 - Google DeepMind - https://www.pexels.com/@googledeepmind - Adis Resic - https://www.pexels.com/@adis-resic-297996969 - K - https://www.pexels.com/@kelly - ArtHouse Studio - https://www.pexels.com/@arthousestudio - Tima Miroshnichenko - https://www.pexels.com/@tima-miroshnichenko - Braeson Holland - https://www.pexels.com/@braeson-holland-3640662 - Emrul Kausar Emon - https://www.pexels.com/@emrulkausar - Dominik Zítka - https://www.pexels.com/@dominik-zitka-911305321 - Diego Castro Calderon - https://www.pexels.com/@diego-castro-calderon-531372879 - ROMAN ODINTSOV - https://www.pexels.com/@roman-odintsov - The Instagrapher - https://www.pexels.com/@theinstagrapher - Usman AbdulrasheedGambo - https://www.pexels.com/@theonlyabdulla - Pavel Danilyuk - https://www.pexels.com/@pavel-danilyuk - Philippe WEICKMANN - https://www.pexels.com/@weickmann - Nino Souza - https://www.pexels.com/@ninosouza
Title: Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Models (Jun 2026) Link: https://arxiv.org/abs/2606.02580 Date: June 1, 2026 Summary: This paper introduces SEIG, an agentic framework that reconstructs editable 3D Blender scenes from a single image using a pretrained vision-language model without specialized 3D tools. By decomposing the reconstruction into sequential stages—geometry, materials, composition, and lighting—and employing a generator-verifier loop, it produces structured code suitable for novel-view synthesis, editing, and physics simulation. Key Topics: - Inverse Graphics - Vision-Language Models - 3D Scene Reconstruction - Agentic Frameworks - Blender Chapters: 00:00 - Introduction & Core Insight 01:46 - Defining Inverse Graphics Problems 03:45 - NeRFs vs Symbolic Code 05:45 - Staged Reconstruction Framework 06:50 - Generator-Verifier Feedback Loop 08:41 - Testing Against Baseline Models 10:22 - Why VLMs Outperform Specialists 11:51 - Reconstructing Complex Visual Scenes 13:10 - Spatial Hierarchy and Composition 14:41 - Granular Editing and Relighting 15:41 - Physics Simulation Capabilities 16:13 - Future Robotics Implications Stock video credits: - Nicola Narracci - https://www.pexels.com/@nicola-narracci-157460431 - Sajjad Sabbir - https://www.pexels.com/@sajjad-sabbir-2151690825 - Chandresh Uike - https://www.pexels.com/@chandresh-uike-754623426 - Sasha Poberailo - https://www.pexels.com/@sasha-poberailo-113008664 - Shutter Break - https://www.pexels.com/@shutterbreak - Кирилл Левченко - https://www.pexels.com/@2156561057 - Computer_Scientist - https://www.pexels.com/@computer_scientist-3594007 - cottonbro studio - https://www.pexels.com/@cottonbro - Rafael Minguet Delgado - https://www.pexels.com/@thales13 - The MoonRunners - https://www.pexels.com/@themoonrunnerrs - Ron Lach - https://www.pexels.com/@ron-lach - Nitin Khajotia - https://www.pexels.com/@nitin-khajotia-55235902 - jorguez - https://www.pexels.com/@jorguez-191283684 - Pressmaster - https://www.pexels.com/@pressmaster - Marina Leonova - https://www.pexels.com/@marina-zasorina - Jakub Zerdzicki - https://www.pexels.com/@jakubzerdzicki - Coverr - https://www.pexels.com/@coverr - Thirdman - https://www.pexels.com/@thirdman - KoolShooters - https://www.pexels.com/@koolshooters - Kindel Media - https://www.pexels.com/@kindelmedia - Joshua Malic - https://www.pexels.com/@joshua-malic-25131152 - Life Of Pix - https://www.pexels.com/@life-of-pix - PNW Production - https://www.pexels.com/@pnw-prod - Vladislav Styazhkin - https://www.pexels.com/@vladislav-styazhkin-157457354 - John Smith - https://www.pexels.com/@john-smith-135446067 - Google DeepMind - https://www.pexels.com/@googledeepmind - Adis Resic - https://www.pexels.com/@adis-resic-297996969 - K - https://www.pexels.com/@kelly - ArtHouse Studio - https://www.pexels.com/@arthousestudio - Tima Miroshnichenko - https://www.pexels.com/@tima-miroshnichenko - Braeson Holland - https://www.pexels.com/@braeson-holland-3640662 - Emrul Kausar Emon - https://www.pexels.com/@emrulkausar - Dominik Zítka - https://www.pexels.com/@dominik-zitka-911305321 - Diego Castro Calderon - https://www.pexels.com/@diego-castro-calderon-531372879 - ROMAN ODINTSOV - https://www.pexels.com/@roman-odintsov - The Instagrapher - https://www.pexels.com/@theinstagrapher - Usman AbdulrasheedGambo - https://www.pexels.com/@theonlyabdulla - Pavel Danilyuk - https://www.pexels.com/@pavel-danilyuk - Philippe WEICKMANN - https://www.pexels.com/@weickmann - Nino Souza - https://www.pexels.com/@ninosouza



