- About WOMBO
- Intro to Subnet 30
- Miners and Validators Functionality
- Get Started with Mining or Validating
- Applications
- Roadmap
WOMBO is one of the world’s leading consumer AI companies, and early believers in generative AI.
We've launched two #1 apps - together, they’ve been downloaded over 200M times and have each hit #1 on the app stores in 100+ countries
These results were only possible due to the immense capabilities of bleeding edge generative AI techniques and the power of open source AI. Our unique understanding of this research space through a consumer entertainment lens allows us to craft products people love to use and share.
We are at the very beginning of the Synthetic Media Revolution, which will completely transform how people create, consume, and distribute content. We're building the apps and infrastructure to power this and bring AI entertainment potential to the masses.
Our mission is to spread fun and creative uses of Al to enhance the human experience. This subnet is the foundation of a decentralized content creation and distribution engine that will transform digital media. With an initial focus on generating captivating images featuring TAO symbolology, the subnet is evolving to support a wide array of media formats and modalities.
Subnet 30 generates compelling content, powers real-world applications, and incentivizes digital posting and social sharing. As the subnet expands, it will unlock new opportunities for people to earn subnet emissions directly through their social media performance.
Subnet 30 incentivizes miners and validators to contribute to the generation and validation of high-quality, engaging content through a unique reward mechanism.
The reward mechanism for Subnet 30 scores the initial response based on the average similarity of randomly selected diffusion steps, rather than the final image. This approach reduces the computation required by validators, preventing them from doubling their workload. The score is then multiplied by the requests per second (RPS) the miner can handle and the success rate to adjust the weight based on the miner's error rate.
Miners are incentivized based on their concurrency, which measures their ability to handle production load. This encourages miners to run behind load balancers to increase the number of requests they can process simultaneously, ultimately improving the subnet's overall performance and reliability.
Currently, Subnet 30 primarily uses randomized strings for validation, with a 25% chance of validating real user data submitted through front-end applications. Failing validation on real user data incurs a higher penalty, as it may indicate an attempt to cheat the system. In the future, the subnet plans to replace randomized validation strings with actual prompts sourced from a prompt bank containing over 2 billion entries.
Successful generation for real user requests grants a bonus to miners, while failing to respond to a user request entirely results in a harsh penalty, as it reduces the overall reliability of the subnet.
- Receieve prompts or validation strings from the subnet.
- Generating images or other content using the custom models and content generation tools.
- Return the generated content to the subnet for validation and distribution.
- Generate queries for miners using randomly generated prompts or actual user requests submitted through applications on the subnet.
- Validate the content generated by miners using the provided diffusion similarity scoring API or by running their own instance of the API for increased decentralization.
- Score the validated content based on relevance, novelty, and detail richness.
- Submit batches of validated content to the subnet for distribution and storage.
To start, clone the repository and cd to it:
git clone https://github.com/womboai/wombo-bittensor-subnet
cd wombo-bittensor-subnet- Recommended: GPU with at least 24GB of VRAM
Miners by default use an API for image generation, this can be set up as follows with PM2:
-
Set the python packages up
cd image-generator ./setup.sh -
Then run with PM2
pm2 start run.sh --name wombo-image-generator --interpreter bash
Note that running one image-generator with one GPU is likely to be ineffective, running multiple image generators with a load balancer is likely to get better rewards.
To set the miner neuron up,
-
Set the python packages up
cd miner ./setup.sh -
Then run with PM2, replacing the arguments
pm2 start run.sh --name wombo-miner --interpreter bash -- \ --netuid {netuid} \ --wallet.name {wallet} \ --wallet.hotkey {hotkey} \ --generation_endpoint http://localhost:8001/api/generate \ --blacklist.force_validator_permit
The miner will then run on your network provided the port 8091 is open.
The responsibility of keeping the miner/repo up-to-date falls on you, hence you should git pull every once in a while.
Running a validator with PM2 is similar to running a miner:
- Recommended: GPU with at least 24GB of VRAM
To run with PM2, do the following, replacing the arguments
pm2 start run_autoupdate.sh --name wombo-validator --interpreter bash -- \
--netuid {netuid} \
--wallet.name {wallet} \
--wallet.hotkey {hotkey}The run script will keep your validator up to date, so as everything is properly configured, everything should run properly.
Bottensor enables users to create images featuring the TAO symbol through the BitTensor network.
Soon Subnet 30 will power the WOMBO Dream and WOMBO Me mobile and web apps, with over ~200K DAU, and ~5M MAU.
- Implement and optimize image generation capabilities within the subnet, with telemetry & monitoring
- Power WOMBO's Dream product with Subnet 30's image generation capabilities (millions of images per day)
- Establish manual incentive layer and reward distribution system for miners creating and distributing viral content
- Develop a suite of memetic primitives and models for diverse media creation
- Automate the incentive layer and create an authentication and payout system
- Develop a suite of memetic primitives and models for diverse media creation. Expand media generation capabilities to support various formats and modalities
- Integrate WOMBO's flagship products (Dream and WOMBO Me) to be fully powered by Bittensor
- Implement a two-layered incentive system, rewarding both efficient media generation and real-world virality
- Launch the real-world virality incentive system, rewarding social media performance
- Develop tools and platforms to facilitate community-driven meme creation and sharing
- Expand the reach and user base of WOMBO's products to drive mainstream adoption of Bittensor
- Collaborate with large influencers and content creators to showcase the power of Subnet 30's AI capabilities
- Release more open-source AI pipelines and APIs for generating viral content and creative assets
- Encourage third-party developers to build applications and services on top of Subnet 30's capabilities
The WOMBO Bittensor subnet is released under the MIT License.
Connect with us on social media
