NicheImage - Image Generating Subnet

---

Introduction

Welcome to the NicheImage - A decentralized network of image models powered by the Bittensor protocol.

This README provides an overview of the project's structure and example usage for both validators and miners.

NicheImage Studio ✨ - https://nicheimage.streamlit.app

Bittensor Resources

• Discord
• Network Information
• Bittensor Homepage

Project Structure

• image_generation_subnet: Contains base, feature functions, and utilities for validators and miners.
• neurons: Contains the validator and miner loop.
• dependency_modules: Includes servers for prompt_generation, rewarding, and miner_endpoint.

Installation

1. Clone the repository.

git clone https://github.com/NicheTensor/NicheImage.git

2. Install the dependencies.

cd NicheImage
pip install -r requirements.txt

3. Install the project.

pip install -e .

Example Usage

Before running the following commands, make sure to replace the placeholder arguments with appropriate values.

Start Miner

Before running the following commands, make sure to replace the placeholder arguments with appropriate values.

First you need to start an image generation API on a gpu server that your miners can use. A RTX 3090 GPU is enough for several miners.

python dependency_modules/miner_endpoint/app.py -h
usage: app.py [-h] [--port PORT] [--model_name {RealisticVision,SDXLTurbo,AnimeV3,RealitiesEdgeXL}]

options:
  -h, --help            show this help message and exit
  --port PORT
  --model_name {RealisticVision,SDXLTurbo,AnimeV3,RealitiesEdgeXL}

python dependency_modules/miner_endpoint/app.py --port <port> --model_name <model_name>

You can also run with pm2.

• SDXLTurbo:

pm2 start python --name "image_generation_endpoint_SDXLTurbo" -- -m dependency_modules.miner_endpoint.app --port 10006 --model_name SDXLTurbo

• RealitiesEdgeXL:

pm2 start python --name "image_generation_endpoint_RealitiesEdgeXL" -- -m dependency_modules.miner_endpoint.app --port 10006 --model_name RealitiesEdgeXL

• RealisticVision

pm2 start python --name "image_generation_endpoint_RealisticVision" -- -m dependency_modules.miner_endpoint.app --port 10006 --model_name RealisticVision

• AnimeV3

pm2 start python --name "image_generation_endpoint_AnimeV3" -- -m dependency_modules.miner_endpoint.app --port 10006 --model_name AnimeV3

Then you can run several miners using the image generation API:

pm2 start python --name "miner" \
-- \
-m neurons.miner.miner \
--netuid <netuid> \
--wallet.name <wallet_name> --wallet.hotkey <wallet_hotkey> \
--subtensor.network <network> \
--generate_endpoint <your_miner_endpoint>/generate \
--info_endpoint <your_miner_endpoint>/info \
--axon.port <your_public_port> \

You can also start with pm2, here is an example:

pm2 start python --name "miner" -- -m neurons.miner.miner --netuid 23 --wallet.name <wallet_name> --wallet.hotkey <wallet_hotkey> --subtensor.network finney --generate_endpoint http://127.0.0.1:10006/generate --info_endpoint http://127.0.0.1:10006/info --axon.port 10010

View logs

pm2 logs miner

Start Validator

Requirements: A validator only needs a cpu server to validate by using our free to use APIs for checking image generation. This is the default setting and requires no configuration.

However, it is possible to run your own image checking APIs if you prefer. This does require a GPU with min 20 GB of ram. You can see how to do this here.

If validators opt in to share their request capacity they will get paid for each image they generate. Opt in is done by specifying --proxy.port If passed, a proxy server will start that allows us to query through their validator, and the validator will get paid weekly for the images they provide.

Start Validator with Default Settings

pm2 start python --name "validator_nicheimage" \
-- -m neurons.validator.validator \
--netuid <netuid> \
--wallet.name <wallet_name> --wallet.hotkey <wallet_hotkey> \
--subtensor.network <network> \
--axon.port <your_public_port> \
--proxy.port <other_public_port> # Optional, pass only if you want allow queries through your validator and get paid

View logs

pm2 logs validator_nicheimage

Schedule update and restart validator

Pull the latest code from github and restart the validator every hour. Notice, the validator must be named validator_nicheimage for the auto-updates to restart the process, so do not change the name from validator_nicheimage.

pm2 start auto_update.sh --name "auto-update"

Multi-GPU Usage

Set CUDA_VISIBLE_DEVICES in your environment to specify which GPU (by index) the model should be deployed to.

export CUDA_VISIBLE_DEVICES=1

Or if you are using pm2

CUDA_VISIBLE_DEVICES=1 pm2 start .....

Roadmap

We will release updates on Tuesdays, in order to make it predictable for when changes to the network will be introduced. Furhter we will do our best to share updates in advance.

Here is the current roadmap for the subnet:

13 Feb: Upgrade model - SDXL Turbo will be gradually replaced by another high quality SDXL model

27 Feb: Adding Img2Img and ControlNet, updates to NicheImage Studio

March: Adding MidJourney to the network- one of the most advanced image generation engine

April: Adding open categories - where different models can compete and the best ones will win and remain on the network