Model Deployment Flow

Model Commitment Generation

The creator prepares a cryptographic commitment to the model configuration, including:

• Base model reference (e.g., stable-diffusion-v1-5) • LoRA weights hash • Training data hash (Merkle root, optional for large datasets)

The model commitment is computed as: modelCommitment = keccak256(baseModelReference ∥ loraWeightsHash ∥ trainingDataHash)

Note: For large training datasets, the training data hash may be omitted or computed over a representative subset to maintain practical deployment costs.

Deterministic Model Contract Deployment

The Model Deployment Contract (MDC) uses CREATE2 to deploy a unique smart contract for the model

• The model commitment is used as the salt for CREATE2, ensuring that any change in model configuration results in a new contract address:

_{modelAddress = CREATE2 (MDC, modelCommitment, initCode)}

• The deployed contract stores the model commitment and relevant metadata

On-Chain Identity Binding

The deployed contract stores: • creator (immutable externally owned account) • modelCommitment (as above) • timestamp (block number or timestamp)

Registry Integration

• The contract address is registered with the Model Factory • All downstream minting and proof verification reference this model contract, ensuring outputs are traceable and verifiable

This design ensures that all AIGC tokens minted through Taste can be cryptographically linked to their generating model and — by extension — the model’s creator. It also enables future possibilities for composable model marketplaces, DAO-curated inference engines, and decentralized attribution systems.