Hi3D v1.0 · 3D refinement on Apple silicon
Collection
Public MLX inference, selective FP16 / INT8 / INT4 / experimental W8A8 downloads, shared components and measured mesh comparisons. • 7 items • Updated
Hi3Dv1.0-mlx · shared components
← Project & inference runtime · Results
The shared FP32 VAE (0.35 GiB), scheduler, pipeline index and transformer configuration files. Download once and reuse with FP16, INT8, INT4 or experimental W8A8 transformers. This repository does not contain the transformer weights.
Install the inference package from the main project, then run:
hi3d-mlx download --variant fp16 --resolution 512
The downloader selects the required files across repositories and verifies SHA-256 digests. A 512³ run requires only transformer_512; 1024³ and 1536³ require both transformers. Image encoders download separately at pinned revisions.
These shared checkpoint bytes are unchanged from the upstream release. Hi3D and third-party terms are preserved in LICENSE, NOTICE and THIRD_PARTY_LICENSES.md.
Hardware compatibility
Log In to add your hardware
Quantized
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for devin-lai/Hi3Dv1.0-mlx-shared
Base model
hi3d-official/Hi3Dv1.0