NeuralAI

---

3D Asset Generator

Discord • Website • App

---

• Introduction
• Key Features
• Miner and Validator Functionality
  • Miner
  • Validator
• Installation
  • Running Miners
  • Running Validators
• Benefit and use cases
  • Benefits
  • Use Cases
• Upgrade

Introduction

NeuralAI is a Bittensor subnet dedicated to the generation of 3D models using advanced neural network techniques. Our goal is to allow developers and artists access to tools that simplify the creation of high-quality 3D assets for various applications, including gaming, virtual reality, and simulations.

Key Features

• High-Quality 3D Generation: Generate realistic 3D models from simple inputs.
• Customizable Parameters: Adjust settings to fine-tune the output according to specific project needs.
• Real-Time Rendering: View generated models in real-time with minimal latency.
• Cross-Platform Compatibility: Easily integrate with popular game engines and 3D modeling software.

Miner and Validator Functionality

Miner

• Miners in the NeuralAI Subnet are responsible for powering the computational infrastructure that enables the 3D asset generation and processing capabilities.
• Generally, miners queue and when receiving a request (the format will be text or image format) from validators, they produce a 3D mesh object by contributing their computational resources.
• After producing a 3D mesh object, the object is sent to validators who assess and provide scores that evaluate the quality of the production and validators announce the scores to miners about their quality and speed of work.

Validator

• Validators are responsible for verifying and validating miners’ work regarding the quality of 3D mesh generation within the network
• First validators assign tasks to available miners by sending prompt made of text or image format.
• When receive the miner’s work, they render images in different axis from the response and validate the scores by comparing the similarity of the prompt and rendered images.
• Validators validate scores based on the quality of work and execution time.
• After validating the scores, they set miners’ weights and send scores to miners so that miners are able to recognize their performance of models.

Installation

Running Miners

We recommend setup using Python>=3.10, PyTorch>=2.1.0, and CUDA>=12.1.

1. First, download the source from the github:

git clone https://github.com/GoNeuralAI/neural-subnet/
cd neural-subnet

2. Install pm2 if you don't have it pm2.io.

sudo apt install npm && sudo npm install pm2 -g && pm2 update

3. Then install the Neural package: pip install -e .

Run Miner with PM2

pm2 start python3 neurons/miner.py --netuid {netuid} --wallet.name {wallet} --wallet.hotkey {hotkey} --axon.port {port} --axon.external_port {port} # if you have both interal and external port

Generation endpoint with PM2

cd generate

# create conda env
conda create -n generation python=3.9
conda activate generation

bash env_install.sh

# Then download the model using the following commands

mkdir weights
huggingface-cli download tencent/Hunyuan3D-1 --local-dir ./weights

mkdir weights/hunyuanDiT
huggingface-cli download Tencent-Hunyuan/HunyuanDiT-v1.1-Diffusers-Distilled --local-dir ./weights/hunyuanDiT


pm2 start "python3 serve.py --port 8093" --name generation_endpoint #{default local port 8093}

Running Validators

1. First, download the source from the github:

git clone https://github.com/GoNeuralAI/neural-subnet/
cd neural-subnet

2. Install pm2 and jq package if you don't have it pm2.io.

sudo apt update && sudo apt install jq
sudo apt install npm && sudo npm install pm2 -g && pm2 update

3. Then install the Neural package: pip install -e .

Validation endpoint with PM2

cd validation

pip install git+https://github.com/openai/CLIP.git
pip install git+https://github.com/facebookresearch/pytorch3d.git
(if you get an issue with this, please run the command below
apt-get install python3-dev)

pip install -r requirements.txt

Run Validator with PM@

pm2 start run.sh --name neural_v_autoupdate -- --netuid {46} --wallet.name {wallet} --wallet.hotkey {hotkey} --axon.port {port} --logging.debug

Benefit and use cases

Benefits

• Quality-Focused Approach By prioritizing the quality of the generated 3D objects over the speed of generation, the subnet can ensure that the output meets the desired high standards. This focus on quality helps to maintain the integrity and usability of the 3D assets, which is crucial for applications that require precise and visually appealing 3D content.
• Efficient Computational Resource Allocation The 'Validator First' structure allows the validators to actively assign computational work to the miners, optimizing the utilization of available resources. This approach can help avoid potential issues related to validator traffic and heavy computational workloads, ensuring a more stable and reliable operation of the subnet.
• Outlier Detection and Filtering The use of Isolation Forest to detect and filter out outliers in the image-text similarity scores helps to improve the overall quality of the generated 3D objects. By removing outliers, the subnet can focus on the most relevant and high-quality associations between the images and text prompts, leading to better 3D model generation.

Use Cases

(Usecase part should be included in the subnet documentation in the subnet github repository.)

• 3D Asset Generation for Content Creation The subnet can be used to generate high-quality 3D assets for various content creation applications, such as video games, virtual reality experiences, architectural visualizations, and product design. The focus on quality ensures that the generated 3D models can be seamlessly integrated into these applications, providing realistic and visually appealing content.

• 3D Modeling Assistance for Designers and Creators The subnet can be integrated into design tools and platforms, providing designers and creators with a powerful AI-driven 3D modeling assistance. Users can leverage the text-to-3D capabilities to quickly generate 3D models based on their descriptive prompts, accelerating the content creation process.

• Educational and Training Applications The subnet can be utilized in educational and training environments, where the generation of accurate 3D models can aid in visualization, simulation, and interactive learning experiences. The high-quality 3D assets produced by the subnet can enhance the educational value and immersive nature of these applications.

• Prototyping and Visualization for Product Development The subnet can be employed in product development workflows, enabling designers, engineers, and product teams to quickly generate 3D prototypes and visualizations based on their conceptual ideas. This can streamline the iterative design process and facilitate more informed decision-making during the product development lifecycle.

By focusing on quality, efficient resource utilization, and outlier detection, the proposed subnet design can be a valuable asset in various industries and applications that require high-quality 3D content generation and modelling capabilities.

Upgrade

• Version 1 focuses on validating the quality of miner responses in an acceptable duration. validators Validate miners with “Challenge” synapse only.

• Version 2 will extend to validate the miners’ generation capacity as well as quality. In Version 1, it covers only text to 3d generation, we will update image to 3d as well and integrate with our end-user app when the quality of miner generation results will rise and better than our edge of our app.

Further subnet roadmap is in progress.