Subnet 26
Tensor Alchemy
Tensor Alchemy
TensorAlchemy is a AI-powered image generation subnet, merging decentralized tech with digital art creation
SN26 : Storb
| Subnet | Description | Category | Company |
|---|---|---|---|
| SN26 : Storb | Storage | Storage | Storb |
TensorAlchemy is Bittensor subnet 26’s innovative image generation system, utilizing decentralized AI technology to transform visual content creation. They harness the Bittensor network to integrate advanced machine learning models, fostering a collaborative and transparent environment for artists and developers. This initiative marks a significant advancement in blending AI with artistic expression, pushing the boundaries of digital art and image synthesis with its unique approach and proprietary human scoring system.
Subnet 26 underwent around 8 weeks of development before its launch without prior testing on a test net. The launch unfolded smoothly, followed by integrating a Discord bot for user engagement and image generation within the subnet.
Decentralized Innovation
Powered by the decentralized Bittensor network, TensorAlchemy taps into the global network of nodes for scalability, security, and limitless creative potential.
Human-Centric AI
Central to TensorAlchemy is their pioneering human scoring system. By integrating human feedback and guidance, they enhance and optimize AI outputs, pushing the boundaries of machine-generated art.
Incentivizing Engagement
They believe in rewarding community contributions. Through their SAAS business model, participants can engage in AI training datasets and earn cryptocurrency by voting on image rounds. Join them in shaping the future of AI-driven creativity!
Alchemy Studio
Explore the capabilities of TensorAlchemy through Alchemy Studio, their intuitive web-based platform. Featuring a user-friendly interface and comprehensive tools, Alchemy Studio facilitates exploration, creation, and sharing of AI-generated images.
Model Usage
Users tend to use the stock model, resulting in a tight grouping and satisfied performance levels. Some users opt to build and optimize their own models for higher scores, contributing to a varied network. The majority relying on stable diffusion leads to less spread compared to diverse model usage. Utilizing the stock stable diffusion model involves upgrading miner software post-release of new models. Determining the number of steps and resolution in model execution significantly impacts response times and image quality.
Validators and Miners
Validators generate prompts using Open AI or Corel, sending them to miners for image creation. Corel integration enhances prompt generation options, providing flexibility in choosing AI sources. Initiating manual validation introduces a weightage system, with an image reward model and human input contributing to miner incentives. Phased implementation includes validator-exclusive access in phase one, expanding to users in subsequent phases for broader participation. Clear communication with both miners and validators is a priority to ensure announcements and updates are easily visible, keeping them informed about important information or future plans.
Image Selection and Reward Model
The process of selecting the best image involves choosing one from a batch of 12 images within a short time frame, currently about 25 seconds. There are discussions on varying mechanisms such as user ordering or selecting the top images from a batch to optimize data collection. A key focus is on developing a custom image reward model fine-tuned based on preferential data to guide the network towards continuously producing higher-quality images. The goal is to continually optimize and enhance image generation through this refined reward system.
The team’s mutual interest in AI and image generation led to the inception of Tensor Alchemy. Emperor Wang from the Tensor Alchemy team has a history of programming since a young age and a self-employed background in web development and AI interest, Emperor Wang’s technical skills were pivotal in preparing to launch Subnet 26. Python proficiency, garnered through exploring AI and machine learning applications, equipped him to tackle the challenges of the subnet launch.
Within the team, there is a Full Stack Engineer with over 7 years of industry experience, dedicated to developing TensorAlchemy Studio. With a strong passion for creating intuitive user experiences and robust architectures, they are instrumental in bringing Subnet 26 innovative ideas to fruition.
There is also a Python Engineer with over 6 years of industry experience, focused on developing the backbone of their services by working on backend APIs. Their expertise in Python and system integration ensures subnet 26 solutions will operate seamlessly and efficiently.