ERC-7007 is an Ethereum token standard for verifiable AI-generated content. Proposed in May 2023, it extends ERC-721 with onchain interfaces that cryptographically link a token to the AI model, prompt, and output that produced it. Every mint carries a proof, not just metadata.
​

The standard answers a practical question: how do you prove that a specific piece of content came from a specific model using a specific prompt? AI image generators, music composers, and text tools now produce work indistinguishable from human output. ERC-7007 turns that output into a token whose provenance can be checked by any contract on the network.
​

For developers building across chains, ERC-7007 matters because it shapes how AI-generated assets behave inside wallets, marketplaces, and cross-chain infrastructure. The standard is also becoming a foundational primitive for agentic commerce, where autonomous AI agents need machine-readable, settlement-ready assets they can transact in onchain. This article walks through how ERC-7007 works, how verification is implemented, and where the standard fits in the broader agentic-commerce stack.
​

ERC-7007 extends ERC-721 by adding three primitives: prompt-indexed token IDs, onchain verification interfaces, and a standardized AIGC metadata schema. Every ERC-7007 token is also a valid ERC-721 token, so it works with existing wallets and marketplaces. The added interfaces let a contract prove that a token's content was generated by a named model from a specific prompt.
​

The ERC-721 standard defines the interface for non-fungible tokens on Ethereum. Each token is unique, identified by a token ID, and can be owned, transferred, and approved for third-party management. It is the backbone of nearly every NFT marketplace and collection in existence.
​

ERC-7007 builds directly on top of ERC-721. Any ERC-7007 token is also a valid ERC-721 token, which means it works with existing wallets, marketplaces, and tooling out of the box. The additions that matter for AI-generated content are:
​

Prompt-indexed token IDs. In a standard ERC-721 collection, token IDs are arbitrary numbers. In ERC-7007, the token ID is derived from the prompt used to generate the content. This creates a deterministic link between what was asked and what was produced, meaning each unique prompt maps to one unique token.
​

Onchain verification interfaces. ERC-7007 introduces an addAigcData function and a verify function. Together, these let a smart contract store and check whether a given output was genuinely produced by a specific AI model using a specific input. This is not just metadata stored in a JSON file. It is a verification mechanism baked into the contract logic.
​

Standardized AIGC metadata schema. The standard defines a JSON schema for AI-generated content metadata that goes beyond the typical NFT name-description-image triplet. It includes fields for the model used, the prompt submitted, and the verification proof, giving collectors and platforms structured data to work with.
​

These additions mean that an ERC-7007 token does not just represent ownership of a digital asset. It represents a verifiable claim about how that asset was created.
​

ERC-7007 supports two verification methods: zero-knowledge machine learning (zkML) and optimistic machine learning (opML). zkML produces a cryptographic proof at mint time that attests the output came from the named model and input. opML publishes the output and opens a challenge window in which anyone can dispute it. Both feed the same verifier interface.
​

In the zkML flow, an AI model is published onchain along with a verifier contract. When a user submits a prompt, the model produces an output. Simultaneously, a zero-knowledge proof is generated that attests the output was correctly derived from the model and the input, without revealing the model's internal weights or architecture.
​

The proof is then submitted to the verifier contract. If it checks out, the content is minted as an ERC-7007 NFT. The model owner's intellectual property stays protected, yet anyone can independently verify the output's authenticity. ORA's reference implementation documents the proving pipeline in detail.
​

The opML approach takes a different tack. Instead of generating a cryptographic proof upfront, the system operates on a trust-but-verify basis. The AI-generated output is published, and there is a challenge period during which anyone can dispute the correctness of the inference. If no successful challenge occurs within the window, the output is considered valid.
​

This is conceptually similar to how optimistic rollups handle transaction finality on Ethereum, where results are assumed correct unless someone proves otherwise. The trade-off: opML is cheaper and faster to produce than a full zero-knowledge proof, but it requires a challenge period before the content can be considered fully verified.
​

ERC-7007 accommodates both methods through its modular verifier interface, letting developers choose the approach that best fits their use case.
​

An ERC-7007 deployment has three core parts: an AIGC-NFT contract compliant with ERC-721 and ERC-7007, a separate verifier contract implementing the verify function, and optional enumerable or updatable extensions. The verifier is decoupled so different collections can plug in zkML, opML, or future proving systems without redeploying the NFT layer.
​

The AIGC-NFT smart contract is the core contract, compliant with both ERC-721 and ERC-7007 interfaces. It handles minting, ownership, transfers, and the AIGC-specific data storage.
​

The verifier smart contract implements a verify function. When given an inference task and its corresponding proof (either a ZK proof or an opML finalization), it returns a boolean result. This contract is separate from the NFT contract itself, which means different collections can plug in different verification mechanisms.
​

Optional extensions include an enumerable interface that maps token IDs to prompts (and vice versa), making it possible to look up the full prompt history of a collection. There is also an updatable extension for opML scenarios, where the AIGC data might change during the challenge period if a dispute succeeds.
​

The specification also addresses frontrunning. Because prompts are submitted onchain during minting, an attacker could observe a pending transaction and submit the same prompt first. The EIP recommends commit-reveal schemes or time-locks to prevent this.
​

For developers working with EVM-compatible chains and cross-chain token standards, ERC-7007's backward compatibility with ERC-721 means these tokens can move through existing bridges and marketplaces without custom integration work.
​

ERC-7007 turns every AI inference into a verifiable, settlement-ready asset that an autonomous agent can pay for onchain. Because the token ID is derived from the prompt and the verifier contract attests the output, an agent can treat the mint receipt as a machine-readable invoice and route a stablecoin payment to the model creator without trusting a centralized billing endpoint.
​

A concrete walkthrough: a research agent needs an image generated from a specific prompt. The agent submits the prompt to an ERC-7007-compliant model contract, the model produces the output and a zkML proof, and the contract mints the token to the agent's address. In the same transaction, or via an ERC-7683 intent, the agent settles the license fee in USDC. The model creator's wallet receives the payment on whichever chain they specified, and the agent walks away with a token whose provenance any auditor can later check.
​

This pattern matters because it removes humans from the loop without removing accountability. The verifier contract proves the model produced the output, the ERC-7007 token records the prompt-to-output binding, and the stablecoin transfer settles the economic relationship. With USDC at $75.6B in circulating supply and USDT at $187.2B as of Q2 2026 (DeFiLlama), stablecoin liquidity is deep enough to underwrite high-frequency agent-to-model payments at production scale.
​

ERC-7007 is the first token standard that turns every AI inference into a verifiable, settlement-ready asset an autonomous agent can pay for onchain. It is not just provenance for AI art. It is the rails for machine commerce over AI work.
​

ERC-7007's prompt-indexed token IDs and verifier contract enable per-inference micropayment loops between agents and models. Each prompt produces a uniquely identified token whose mint can be gated on a stablecoin transfer, turning every inference into a discrete, audited transaction. The pattern replaces opaque API-key billing with onchain settlement at the granularity of a single model call.
​

Under the API-key model, an agent authenticates once and is billed in aggregate at the end of a period. Disputes require log access from the provider. Under the ERC-7007 model, the contract is the receipt: each mint records the prompt, the model identifier, the verification proof, and the timestamp. If the agent's accounting disagrees with the model creator's, both parties can replay the verifier against the same data.
​

The economics shift too. API-key billing favors providers who can support large minimum commits. Per-inference settlement opens the door to long-tail models that price by the call, paid in stablecoins the agent already holds. A model creator can publish a contract that charges 0.05 USDC per generation and accept payment from any wallet on a supported chain. Tooling like ORA's onchain AI Oracle demonstrates the loop in production.
​

An ERC-7007 mint can happen on one chain while the model creator receives payment on another. Pairing ERC-7007 with cross-chain intent standards like ERC-7683 lets agents post a settlement intent in their preferred currency, and solvers fulfill the payment to the creator on their preferred chain. The token's provenance data and the settlement leg travel through different rails but stay logically linked.
​

This matters because model creators and agents rarely operate on the same chain. A model owner might prefer USDC on Base (currently $3.9B TVL per DeFiLlama) for low fees, while an agent operates with treasury on Ethereum ($37.1B TVL) or Solana ($4.8B TVL). Without cross-chain settlement, every model integration forces both parties onto the same network. With ERC-7683 intents executed by solvers, the agent expresses what it wants and the routing layer handles the rest.
​

Eco Routes is one implementation of this intent pattern, settling stablecoin transfers across EVM chains in seconds. The point is not which router wins. The point is that ERC-7007 + intent-based routing together make AI-agent settlement chain-agnostic, which is the property agentic commerce needs to scale beyond single-network demos.
​

ERC-7007 fits any application where the relationship between an AI model, its input, and its output needs to be provable onchain. The most cited examples are AI art, AI music, revenue-sharing for model authors, and data provenance for research. An emerging category is agent-to-agent licensing, where one autonomous agent pays another for verified model output.
​

An artist deploys an AI model onchain. Collectors submit prompts during minting, generating unique pieces. Each artwork is verified as the output of that model with that prompt, and the proceeds flow to the artist. This creates a new model for digital content ownership where the artist monetizes the model itself, not just individual outputs.
​

A musician tokenizes an AI model that generates music tracks. Fans submit prompts or themes, receive personalized compositions, and own the resulting NFTs. The verification layer confirms that each track genuinely came from the artist's model, not a copycat trained on similar data.
​

ERC-7007 enables structures where the original model creator receives ongoing compensation as their model generates content. Because the link between model, prompt, and output is verifiable onchain, revenue-sharing can be automated through smart contract logic rather than relying on honor systems.
​

Beyond creative applications, the standard has potential in any context where proving that a specific AI model produced a specific output matters, from scientific research replication to legal document generation and regulated industries where audit trails are required.
​

AI agents licensing other models' outputs is a fast-emerging use case. A research agent paying an image model per generation can mint ERC-7007 tokens for each output, settle in stablecoins, and retain an onchain audit trail. Downstream consumers of the research can inspect every inference the agent paid for, replay the verifier, and confirm provenance without trusting the agent's word.
​

ERC-7007's strengths are verifiable provenance, ERC-721 backward compatibility, modular verification, and a standardized metadata schema. Its limitations are zkML's computational cost, limited adoption at the Review stage, opML's reliance on active challengers, and the practical impossibility of running full models onchain for most architectures. Most implementations push inference offchain and verify the result.
​

ERC-7007 sits alongside agent-identity standards like ERC-8004 and intent standards like ERC-7683 as the emerging agentic-commerce stack. ERC-7007 covers provenance of AI work, ERC-8004 covers who the agent is, and ERC-7683 covers how the agent settles across chains. Together they let an identified agent transact verified AI output and pay in stablecoins, anywhere.
​

Ethereum's standards family includes ERC-20 for fungible tokens, ERC-721 for non-fungible tokens, ERC-1155 for semi-fungible tokens, ERC-4626 for yield-bearing vaults, and ERC-7683 for cross-chain intent-based transactions. ERC-7007 carves out a specific niche at the intersection of AI and NFTs, adding a verification layer that becomes increasingly relevant as AI-generated content fills digital markets.
​

The agentic-commerce angle is what distinguishes ERC-7007 from prior provenance attempts. Platforms, collectors, and creators all benefit from a standardized way to ask "was this actually made by the model the creator says it was?" Autonomous agents extend that question into a transactional one: "can I pay for this inference, prove I paid for it, and resell or relicense the output later?" ERC-7007 gives the affirmative answer in the form of a token.
​

As stablecoin infrastructure and cross-chain execution layers mature, the ability to move verified AI-generated content tokens across chains with the same ease as any other ERC-721 token becomes a practical reality for both human creators and agent operators.
​

Developers can start with the reference implementation on the EIP page, which includes sample contract code and unit tests. ORA Network provides tooling for deploying AI models onchain with ERC-7007 verification, supporting both zkML and opML proving methods. The community discussion on Ethereum Magicians is the most active forum for implementation questions.
​

Key implementation steps include:

The ERC-7007 discussion thread on Ethereum Magicians remains an active resource for technical questions.
​

This FAQ covers the questions most often asked about ERC-7007, including how it differs from a standard NFT, what zkML and opML actually do, current adoption status, cross-chain portability, supported content types, and how the standard fits into autonomous agent workflows and broader agentic commerce.
​

What is ERC-7007 in simple terms?
​

ERC-7007 is an Ethereum token standard that lets you create NFTs for AI-generated content with built-in proof that a specific AI model produced the content using a specific prompt. It extends the existing ERC-721 NFT standard.
​

How is ERC-7007 different from a regular ERC-721 NFT?
​

A regular ERC-721 NFT represents ownership of a digital asset but says nothing about how the asset was created. ERC-7007 adds verification interfaces that cryptographically prove the content was generated by a particular AI model with a particular input.
​

What are zkML and opML in the context of ERC-7007?
​

zkML (zero-knowledge machine learning) uses cryptographic proofs to verify AI outputs without revealing the model's internals. opML (optimistic machine learning) assumes outputs are correct unless someone successfully challenges them within a time window. Both are supported verification methods within ERC-7007.
​

Is ERC-7007 widely adopted?
​

The standard is in the Review stage as of Q2 2026. Several projects, notably ORA Network, have built implementations, but it has not reached mainstream adoption. Interest is growing as AI-generated content becomes more prevalent and as agentic-commerce use cases create new demand for verifiable AI output.
​

Can ERC-7007 tokens be used on other blockchains?
​

Because ERC-7007 extends ERC-721, these tokens can move to other EVM-compatible chains using standard cross-chain infrastructure. The verification data travels with the token metadata, and intent-based routing standards like ERC-7683 can settle the payment leg on a different chain than the mint.
​

What types of AI content can ERC-7007 handle?
​

The standard is content-agnostic. It can handle images, music, text, video, or any other output that an AI model produces. The verification layer cares about the model-prompt-output relationship, not the content type itself.
​

Can AI agents mint and pay for ERC-7007 tokens autonomously?
​

Yes. An autonomous agent with a wallet can submit a prompt to an ERC-7007 model contract, receive the minted token, and settle the license fee in stablecoins in the same flow. Because the mint is gated on verification and the payment is onchain, no human approval is required at runtime, and the agent's accounting is reproducible by any auditor.
​

How does ERC-7007 fit into agentic commerce?
​

ERC-7007 provides the verifiable-asset layer that agentic commerce needs. Combined with agent-identity standards like ERC-8004 and intent-based settlement standards like ERC-7683, it lets identified agents pay for AI work, prove what they paid for, and route stablecoin payments to model creators across chains. The token is the receipt, the verifier is the auditor, and the intent layer is the router.
​

ERC-7683 Cross-Chain Intents Standard Explained
​Stripe Stablecoin Payments 2026
​x402 Protocol Explained: How AI Agents Pay Onchain
​What Is Agentic Commerce?
​