PartCrafter review
A research-focused local tool for turning one image into semantically distinct 3D parts.
Reviewed by iTechGuides Editors · Editorial team · Updated Oct 2026
PartCrafter is an open-source AI model and implementation for developers and researchers generating structured 3D content from images. Its core capability is image-to-3D generation: one RGB image can produce multiple semantically meaningful mesh parts, decomposable objects, or multi-object scenes. The project uses a compositional latent space with part identity embeddings, and it supports VLM-based automatic part-count suggestion and image style transfer for real-world photos. It runs locally on Windows and Linux.
Its standout feature is part-level generation rather than general-purpose 3D creation. Local-global denoising attention is designed to support both part detail and global coherence, while the compositional representation preserves distinct part identities. That makes PartCrafter a focused option for research workflows involving decomposable objects, structured assets, and scene generation from images. The project also provides pretrained weights, inference scripts, training configurations, and a Hugging Face model repository, giving technical users materials for local inference and further training.
The main trade-off is the implementation's technical setup and narrow scope. Local use requires a CUDA-enabled GPU with at least 8GB of VRAM, so it suits teams with compatible hardware and development capacity more than users seeking a browser-first workflow. Google Gemini is listed as an integration, while text-to-3D generation is not part of the stated capability set. Choose PartCrafter for local, research-oriented image-to-3D work where semantically distinct parts matter; choose another tool if you need hosted SaaS delivery or text-driven generation.
PartCrafter pros and cons
- Where it wins
- Generates multiple semantically distinct 3D parts from one RGB image
- Creates multi-object 3D scenes with local-global attention for coherence
- Includes local inference, training scripts, pretrained weights, and configurations
- Where it doesn't
- Requires a CUDA-enabled GPU with at least 8GB of VRAM
- Supports image-to-3D generation, but not text-to-3D generation
- Focused on local implementation rather than a conventional hosted SaaS product
PartCrafter fact sheet, pricing and score →
Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.
Last updated · How we research and update