InstantMesh
InstantMesh is a free framework that generates a 3D mesh from a single image using the LRM/Instant3D architecture.
What InstantMesh is
InstantMesh is a free framework that generates a 3D mesh from a single image using the LRM/Instant3D architecture. Published by TencentARC, it can run locally or through a self-hosted setup, with Docker also documented. The repository includes a Gradio demo launched with `python app.py` and a `run.py` command-line tool for mesh generation. The default export is an OBJ mesh with vertex colors; an alternate texture-map export takes longer. Image processing uses rembg to isolate the foreground object, though users can skip that step when an input already has an alpha mask. The project provides four sparse-view reconstruction model variants and a customized Zero123++ UNet for white-background image generation. It also includes training code for the reconstruction models and Zero123++ fine-tuning. The makers recommend Python 3.10 or later, PyTorch 2.1.0 or later, and CUDA 12.1 or later; the demo can use two GPUs to save memory. The repository includes an Apache License 2.0. Community implementations are acknowledged for ComfyUI, Replicate, and Google Colab demos.
Good
- Generates a mesh from a single image.
- Includes a Gradio demo and command-line workflow.
- Default OBJ export includes vertex colors.
- Includes four sparse-view reconstruction variants.
- Apache License 2.0.
Watch out
- Requires Python, PyTorch, CUDA, and a compatible GPU.
- Training dataset is not provided.
- Texture-map export takes longer.
Who it suits
InstantMesh may suit developers and technical users who want to create 3D meshes from images in a local or self-hosted environment. The documented setup depends on Python, PyTorch, CUDA, and a compatible GPU.
Verdict
InstantMesh offers local image-to-mesh generation alongside model variants and training code. Its technical requirements make the documented environment an important consideration before setup.
Plans and prices
From InstantMesh's own pricing page, checked 8 Oct 2026.
| Plan | Price | What it gives you |
|---|---|---|
| InstantMesh | Free | Apache-2.0 licensed implementation · local installation requires Python, PyTorch, CUDA, and a compatible GPU |
The facts
Each one read from InstantMesh's own pages, with the date it was checked.
What it does
- Command-line use
- The `run.py` script generates 3D meshes from images via the command line.checked 8 Oct 2026
- Dataset limitation
- The makers state that they cannot provide the training dataset because of its size.checked 8 Oct 2026
- Deployment
- The README documents running the demo locally or with Docker.checked 8 Oct 2026
- Hardware and dependencies
- The makers recommend Python 3.10 or later, PyTorch 2.1.0 or later, and CUDA 12.1 or later; the demo can use two GPUs to save memory.checked 8 Oct 2026
- Image processing
- The project uses rembg to segment the foreground object and supports skipping that step when the input image already has an alpha mask.checked 8 Oct 2026
- Integration
- The README acknowledges community implementations for ComfyUI, Replicate, and Google Colab demos.checked 8 Oct 2026
- License
- The repository includes an Apache License 2.0.checked 8 Oct 2026
- Local demo
- The repository provides a local Gradio demo that can be started with `python app.py`.checked 8 Oct 2026
- Mesh output
- The default export is an `.obj` mesh with vertex colors; an option exports a mesh with a texture map and takes longer.checked 8 Oct 2026
- Model variants
- The project provides four sparse-view reconstruction model variants and a customized Zero123++ UNet for white-background image generation.checked 8 Oct 2026
- Purpose
- InstantMesh is a feed-forward framework for generating a 3D mesh from a single image using the LRM/Instant3D architecture.checked 8 Oct 2026
- Training
- The repository includes training code for sparse-view reconstruction models and Zero123++ fine-tuning.checked 8 Oct 2026
The company
- Founded
- 2024checked 28 Sep 2026
- Maker
- The repository is published by TencentARC; the pages opened do not state its headquarters or founding year.checked 8 Oct 2026
