The Daily Commit · Section Edition Front Page PHP AI Dev EN DE FR ES

TheModelDesk

July 12, 2026
models, agents & local inference

Tooling

Local Image to 3D on Apple Silicon with MLX

Modelr, an open-source macOS/iOS app, brings Hunyuan3D image-to-3D generation to Apple Silicon devices using MLX optimization. Runs in 20–22 seconds on M4 Max (5–7 GB RAM), supports FP16/Q8 quantization for efficient inference without PyTorch overhead. GitHub: ZimengXiong/Modelr.

Developer ZimengXiong has released Modelr, an open-source application for macOS and iOS that enables local image-to-3D generation. The app is based on a Swift-MLX and Python port of Hunyuan3D-Paint and Hunyuan3D-Shape, optimized specifically for Apple Silicon to eliminate the overhead associated with PyTorch or CPU processing.

Performance benchmarks on an M4 Max using FP16 precision show that the small Hunyuan3D shape model generates results in 20.9 seconds with approximately 5.6 GB of RAM, while the large version takes 22.3 seconds with about 7.3 GB of RAM. More intensive paint operations (RGB and PBR) require significantly more time and memory, peaking around 39 GB.

To broaden device compatibility, the model supports Q4 and Q8 quantization, allowing it to run on older Macs and iPhones. The application workflow includes image importing, background removal via SwiftVision, and real-time diffusion and texturing streams.

Both the Modelr app and the underlying weights and source code (available at Hunyuan3D-Swift) are open-sourced on GitHub for developers wishing to integrate fast, low-memory 3D generation into their own Swift applications.

Read the original source ↗

Rate this article: 0

Readers’ Forum

No contributions yet — open the debate.

← The Model Desk — Page C1

Models, agents & local inference · The Daily Commit · Screen edition · Imprint · Privacy Policy