What is Hi3Dv1?
Hi3Dv1, also called Hi3D v1.0, is a geometry-refinement model built on the Sparc3D representation and refinement architecture. It takes an existing coarse mesh and a reference image, then progressively samples sparse latents, decodes them with a sparse VAE, and extracts a more detailed mesh. The official release is an inference project with model weights and documentation; it is not a hosted browser app.
How does the Hi3Dv1 workflow work?
Start with a coarse mesh
The pipeline expects an initial 3D shape. The coarse structure matters because severe errors can remain in the refined result.
Condition on an image
Image features guide sparse geometry refinement around the visible object. Hidden surfaces are plausible completions, not guaranteed scans.
Choose a resolution ladder
The official examples support 512³, 1024³, and 1536³ sparse extraction settings. Higher settings use more passes and more GPU memory.
When is Hi3Dv1 a good fit?
Hi3Dv1 is most useful when you already have a coarse reconstruction and need finer geometric structure for research, asset experiments, or a local 3D pipeline.
Local research and prototyping
Teams with a compatible Linux GPU can inspect the inference code, test their own meshes, and measure refinement quality on a controlled reference set.
Detail-focused asset work
The refinement approach is suited to visible relief, edges, layered surfaces, and other geometric details that a coarse mesh may miss.
Important limits before you run it
Hi3Dv1 generates geometry only: it does not create textures, materials, or vertex colors. The official setup requires Linux, Python 3.10, an NVIDIA GPU with compute capability 8.0 or newer, and a compatible CUDA 12.0–12.5 environment. The 1536³ ladder was validated on a 96 GB NVIDIA H20; feasibility on smaller GPUs is not established. Read the current repository license and branding requirements before a commercial deployment.
Need an online image-to-3D workflow?
If your goal is to turn a product photo or a small set of views into a textured 3D asset without installing a local GPU stack, Trilo3D provides a browser workflow with model settings, visible credit costs, preview, and export options. Trilo3D does not claim to host Hi3Dv1; it is an online alternative for a different workflow.
Hi3Dv1 FAQ
Is Hi3Dv1 the same as Hi3D v1.0?
Yes. Hi3Dv1 is the short name commonly used for the official Hi3D v1.0 release. The official repository is named Hi3Dv1.0.
Does Hi3Dv1 generate textured 3D models?
No. The official scope is geometry refinement. It does not generate textures, materials, or vertex colors, so a separate texturing or material workflow is needed.
Can I run Hi3Dv1 without a local GPU?
The official inference instructions require a Linux NVIDIA GPU with compute capability 8.0 or newer. If you want to test image-to-3D without managing that environment, use a hosted browser workflow such as Trilo3D.
Does Trilo3D run Hi3Dv1?
No. Trilo3D is a separate hosted image-to-3D studio. This page links to the official Hi3Dv1 sources and describes Trilo3D as an online alternative for users who want a simpler workflow.