3DMorph: Single-Image-Guided Local 3D Shape Editing and Morphing

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

3DMorph is a training-free framework for single-image-guided local 3D shape editing and morphing, addressing limitations in existing 3D editing methods. It automatically localizes relevant 3D regions from an edited 2D image and transfers modifications while preserving unmodified areas. The system also generates intermediate shapes for design exploration. A new benchmark, Delta3D, with 74 paired ground-truth edits, was introduced for geometry-centric evaluation. Experimental results show 3DMorph significantly outperforms state-of-the-art generative and editing methods, reducing Chamfer Distance by 19% and increasing IoU by 3 points compared to Trellis-RePaint. It robustly handles diverse 3D modifications and multi-step editing sequences.

Key takeaway

For CAD engineers or 3D modelers seeking efficient local shape modifications, 3DMorph offers a robust, training-free solution. You can now leverage familiar 2D image editing tools to precisely alter 3D mesh geometry, avoiding full model regeneration. This enables rapid design iteration and exploration of intermediate shapes, significantly streamlining workflows for complex engineering assets.

Key insights

3DMorph enables precise local 3D shape editing and morphing from a single edited 2D image without training.

Principles

• Local 3D edits require a two-stage conditioning schedule.
• Hard constraints stabilize global shape during voxel prediction.
• Flow-consistent conditioning refines latent details smoothly.

Method

Detect 2D image differences, project them to a 3D bounding box, apply a two-stage conditioning schedule for voxel prediction and latent refinement, then decode the hybrid Structured Latent (SLAT) into a mesh.

In practice

• Use standard 2D image editing (inpainting, sketching) for 3D modifications.
• Generate intermediate shapes for design-space exploration.
• Perform sequential edits on complex 3D objects.

Topics

• 3D Shape Editing
• Mesh Morphing
• Single-Image Guidance
• Structured Latents
• Delta3D Benchmark
• Generative Models

Code references

• toprei/3dmorph
• black-forest-labs/flux

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• 3DMorph: Single-Image-Guided Local 3D Shape Editing and Morphing — 3DMorph enables intuitive local 3D shape editing from a single image, outperforming existing methods.
• Beyond Voxel 3D Editing: Learning from 3D Masks and Self-Constructed Data — The BVE framework improves 3D editing by using a custom dataset, lightweight modules, and an annotation-free masking strategy.
• Consistent-Inversion: Reverse Consistency Guidance for Structure-Preserving Visual Editing — Reverse consistency guidance corrects structural drift in diffusion-based image editing by checking trajectory reversibility.
• Beyond Prompts: Unconditional 3D Inversion for Out-of-Distribution Shapes — Text-to-3D models can lose prompt sensitivity in "latent sink traps," but retain geometric expressivity.
• Towards 3D-Aware Video Diffusion Models: Render-Free Human Motion Control with Mesh Tokenization — Video diffusion models can achieve 3D-awareness for human motion control by directly using compressed 3D human mesh tokens.
• Native3D: End-to-End 3D Scene Generation via Unified Mesh-Texture Modeling and Semantic Alignment — Direct 3D scene generation via unified mesh-texture modeling and semantic alignment overcomes 2D representation limitations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.