OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Meshxl

MeshXL is a family of pretrained generative models for 3D mesh creation. It uses a Neural Coordinate Field representation to model unstructured mesh data with sequential language-model methods. Its repository includes model weights, inference code and scripts for unconditional generation and partial mesh completion.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Meshxl — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use MeshXL when you want to experiment with pretrained models for unconditional 3D mesh generation or partial mesh completion, rather than start by training a foundation model from scratch.

Who it is for

  • Researchers exploring generative models for 3D mesh data.
  • Developers prototyping 3D asset-generation workflows with pretrained weights.
  • Practitioners working with unconditional mesh generation or partial mesh completion.

Documented uses

  • Generate unconditional 3D mesh samples with a pretrained model.
  • Complete partial 3D meshes.
  • Explore category-specific unconditional generation for chairs, tables, lamps or benches.
  • Study or prototype sequential modeling of unstructured mesh data using a Neural Coordinate Field representation.
Before you choose

Requirements & limits

The public GitHub repository lists an MIT license. The README provides model weights through Hugging Face repositories and documents a local setup script and sample-generation scripts.

  • The model family uses a Neural Coordinate Field representation combining explicit coordinates with implicit neural embeddings.
  • The README states that sample generation defaults to eight GPUs.
  • The base pretrained weights are described as supporting unconditional generation and partial mesh completion; category-specific fine-tuned weights are for unconditional generation.
  • The README marks data preparation and generative pre-training as work in progress.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up MeshXL and prepare pretrained weights

Create the project environment and place pretrained model weights in the repository’s model folders.

Before you start

  • A local MeshXL repository checkout.
  • A shell environment to run the provided setup script.
  • Git LFS for cloning the Hugging Face weight repositories.

Steps

  1. Run the command below from the repository to build the environment.

    Documented command
    bash set_env.sh

  2. Change into ./mesh-xl and clone one or more listed model repositories with the command below, such as https://huggingface.co/CH3COOK/mesh-xl-125m.

    Documented command
    git lfs clone

  3. Place or replace the corresponding pytorch_model.bin file in the matching model folder as described in the README.

Expected outcome

The environment setup script has been run and selected pretrained weights are placed in the corresponding model folders.

Guide limitations

  • The README marks data preparation and generative pre-training as work in progress.
  • This guide covers setup and weight placement; it does not establish that a sample-generation run has succeeded.
Documentation 1 ↗
Source-based practical guide

Run a documented mesh-sampling script

Use one of the repository’s listed scripts to request mesh samples from a pretrained model.

Before you start

  • MeshXL environment and pretrained weights prepared as described in the repository.
  • A machine configured with eight GPUs, the README’s default sample configuration.

Steps

  1. From the repository, choose a model-size script: the command below, the command below or the command below.

    Documented command
    bash scripts/sample-125m.sh
    Documented command
    bash scripts/sample-350m.sh
    Documented command
    bash scripts/sample-1.3b.sh

  2. The scripts use the README’s stated top-k and top-p sampling strategy for diverse samples.

Expected outcome

The selected sample-generation script is invoked for its corresponding model size.

Guide limitations

  • The README gives eight GPUs as the default configuration and does not document alternative hardware requirements.
  • The repository’s sample commands do not specify prompts; the described base-model use is unconditional generation.
  • This is documentation-based guidance, not a verification that the scripts run successfully.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
MeshxlGenerate unconditional 3D mesh samples with a pretrained model.CloudThe public GitHub repository lists an MIT license. The README provides model weights through Hugging Face repositories and documents a local setup script and sample-generation scripts.
Pulse Of MotionEstimate a video's physical frame rate from its visual dynamics.LinuxThe repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence.
Neural Image AssessmentTrain the model using AVA images and rating annotations.Cloud, Linux, macOS, Windows
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
OpenMeshLab/MeshXL · Official homepage title · retained in its source language
How we review sourcesSuggest a correction