> ## Documentation Index > Fetch the complete documentation index at: https://docs.cosmos.network/llms.txt > Use this file to discover all available pages before exploring further. # Non-Fungible Token Transfer > This standard document follows the same design principles of [ICS 20](../ics-020-fungible-token-transfer) and inherits most of its content therefrom, while replacing `bank` module based asset tracking logic with that of the `nft` module. ## Synopsis This standard document specifies packet data structure, state machine handling logic, and encoding details for the transfer of non-fungible tokens over an IBC channel between two modules on separate chains. In this document, `class`, `collection` and `contract` are used interchangeably. The state machine logic presented allows for safe multi-chain `classId` handling with permissionless channel opening. This logic constitutes a *non-fungible token transfer bridge module*, interfacing between the IBC routing module and an existing asset tracking module on the host state machine, which could be either a Cosmos-style native module or a smart contract running in a virtual machine. ### Motivation Users of a set of chains connected over the IBC protocol might wish to utilize a non-fungible token on a chain other than the chain where the token was originally issued -- perhaps to make use of additional features such as exchange, royalty payment or privacy protection. This application-layer standard describes a protocol for transferring non-fungible tokens between chains connected with IBC which preserves asset non-fungibility, preserves asset ownership, limits the impact of Byzantine faults, and requires no additional permissioning. ### Definitions The IBC handler interface & IBC routing module interface are as defined in [ICS 25](../../core/ics-025-handler-interface) and [ICS 26](../../core/ics-026-routing-module), respectively. ### Desired Properties * Preservation of non-fungibility (i.e., only one instance of any token is *live* across all the IBC-connected blockchains). * Permissionless token transfers, no need to whitelist connections, modules, or `classId`s. * Symmetric (all chains implement the same logic, no in-protocol differentiation of hubs & zones). * Fault containment: prevents Byzantine-creation of tokens originating on chain `A`, as a result of chain `B`'s Byzantine behavior. ## Technical Specification ### Data Structures Only one packet data type is required: `NonFungibleTokenPacketData`, which specifies the class id, class uri, class data, token id array, token uri array, token data array, sending address, and receiving address. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} interface NonFungibleTokenPacketData { classId: string classUri: string classData: string tokenIds: string[] tokenUris: string[] tokenData: string[] sender: string receiver: string memo: string } ``` `classId` is a required field that MUST never be empty, it uniquely identifies the class/collection/contract which the tokens being transferred belong to in the sending chain. In the case of an ERC-1155 compliant smart contract, for example, this could be a string representation of the top 128 bits of the token ID. `classUri` is an optional field which, if present, MUST be non-empty and refer to an off-chain resource that is typically a JSON file containing the [class metadata](https://docs.opensea.io/docs/contract-level-metadata); this could be extremely beneficial for cross-chain interoperability with NFT marketplaces like OpenSea. `classData` is an optional field which, if present, MUST be non-empty and contain on-chain class metadata such as royalty related parameters. `tokenIds` array is a required field that MUST have a size greater than zero and hold non-empty entries that uniquely identify tokens (of the given class) that are being transferred. In the case of an ERC-1155 compliant smart contract, for example, a `tokenId` could be a string representation of the bottom 128 bits of the token ID. `tokenUris` array is an optional field which, if present, MUST have the same size as `tokenIds` and hold non-empty entries each of which refers to an off-chain resource that is typically an immutable JSON file containing metadata associated with the token identified by the corresponding `tokenIds` entry. `tokenData` array is an optional field which, if present, MUST have the same size as `tokenIds` and hold non-empty entries each of which contains on-chain application data associated with the token identified by the corresponding `tokenIds` entry. Both `tokenData` entries and `classData` MUST be Base64 encoded strings which SHOULD have the following JSON structure: ```json theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} { "key1" : { "value":"...", "mime":"..." }, "key2" : { "value":"...", "mime":"..." }, ... } ``` `mime` is an optional property that specifies the media type of the corresponding key-value. If a key-value is of the default type of string, then `mime` can be omitted. Otherwise, `mime` MUST be non-empty and have a value that comes from this [list](https://www.iana.org/assignments/media-types/media-types.xhtml). Chain applications are advised to namespace the keys; to achieve maximum interoperability across applications, standardization of these namespaces is desired but out of the scope. An example of `classData` content (raw JSON before being Base64 encoded) is shown below: ```json theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} { "opensea:name" : { "value":"Crypto Creatures" }, "opensea:image" : { "value":"...(Base64 encoded media binary)", "mime":"image/png" }, "opensea:seller_fee_basis_points" : { "value":"100" } } ``` The optional `memo` field is not used within the transfer, however, it may be used either for external off-chain users (i.e. exchanges) or for middleware wrapping transfer that can parse and execute custom logic on the basis of the passed-in memo. If the memo is intended to be parsed and interpreted by higher-level middleware, then these middlewares are advised to namespace their additions to the memo string so that they do not overwrite each other. Chains should ensure that there is some length limit on the entire packet data to ensure that the packet does not become a DOS vector. However, these do not need to be protocol-defined limits. If the receiver cannot accept a packet because of length limitations, this will lead to a timeout on the sender side. As tokens are sent across chains using the ICS-721 protocol, they begin to accrue a record of channels across which they have been transferred. This record information is encoded into the `classId` field. An ICS-721 token class is represented in the form `{ics721Port}/{ics721Channel}/{classId}`, where `ics721Port` and `ics721Channel` identify the channel on the current chain from which the tokens arrived. If `{classId}` contains `/`, then it must also be in the ICS-721 form which indicates that the tokens have a multi-hop record. Note that this requires that the `/` (slash character) is prohibited in non-IBC token `classId`s. A sending chain may be acting as a source or sink zone. When a chain is sending tokens across a port and channel which are not equal to the last prefixed port and channel pair, it is acting as a source zone. When tokens are sent from a source zone, the destination port and channel will be prefixed onto the `classId` (once the tokens are received) adding another hop to the tokens record. When a chain is sending tokens across a port and channel which are equal to the last prefixed port and channel pair, it is acting as a sink zone. When tokens are sent from a sink zone, the last prefixed port and channel pair on the `classId` is removed (once the tokens are received), undoing the last hop in the tokens record. For example, assume these steps of transfer occur: A -> B -> C -> A -> C -> B -> A 1. A(p1,c1) -> (p2,c2)B : A is source zone. `classId` in B: 'p2/c2/nftClass' 2. B(p3,c3) -> (p4,c4)C : B is source zone. `classId` in C: 'p4/c4/p2/c2/nftClass' 3. C(p5,c5) -> (p6,c6)A : C is source zone. `classId` in A: 'p6/c6/p4/c4/p2/c2/nftClass' 4. A(p6,c6) -> (p5,c5)C : A is sink zone. `classId` in C: 'p4/c4/p2/c2/nftClass' 5. C(p4,c4) -> (p3,c3)B : C is sink zone. `classId` in B: 'p2/c2/nftClass' 6. B(p2,c2) -> (p1,c1)A : B is sink zone. `classId` in A: 'nftClass' The acknowledgement data type describes whether the transfer succeeded or failed, and the reason for failure (if any). ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} type NonFungibleTokenPacketAcknowledgement = | NonFungibleTokenPacketSuccess | NonFungibleTokenPacketError interface NonFungibleTokenPacketSuccess { // This is binary 0x01 base64 encoded success: "AQ==" } interface NonFungibleTokenPacketError { error: string } ``` Note that both the `NonFungibleTokenPacketData` as well as `NonFungibleTokenPacketAcknowledgement` must be JSON-encoded (not Protobuf encoded) when serialized into packet data. The non-fungible token transfer bridge module maintains a separate escrow address for each NFT channel. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} interface ModuleState { channelEscrowAddresses: Map } ``` ### Sub-protocols The sub-protocols described herein should be implemented in a "non-fungible token transfer bridge" module with access to the NFT asset tracking module and the IBC routing module. The NFT asset tracking module should implement the following functions: ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function CreateOrUpdateClass(classId: string, classUri: string, classData: string) { // creates a new NFT Class identified by classId // if classId already exists, app logic may choose to update class metadata accordingly } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function Mint(classId: string, tokenId: string, tokenUri: string, tokenData: string, receiver: string) { // creates a new NFT identified by // receiver becomes owner of the newly minted NFT } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function Transfer(classId: string, tokenId: string, receiver: string, tokenData: string) { // transfers the NFT identified by to receiver // receiver becomes new owner of the NFT // if tokenData is not empty, app logic may choose to update token data accordingly } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function Burn(classId: string, tokenId: string) { // destroys the NFT identified by } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function GetOwner(classId: string, tokenId: string) { // returns current owner of the NFT identified by } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function GetNFT(classId: string, tokenId: string) { // returns NFT identified by } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function GetClass(classId: string) { // returns NFT Class identified by classId } ``` #### Port & channel setup The `setup` function must be called exactly once when the module is created (perhaps when the blockchain itself is initialised) to bind to the appropriate port (owned by the module). ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function setup() { capability = routingModule.bindPort("nft", ModuleCallbacks{ onChanOpenInit, onChanOpenTry, onChanOpenAck, onChanOpenConfirm, onChanCloseInit, onChanCloseConfirm, onRecvPacket, onTimeoutPacket, onAcknowledgePacket, onTimeoutPacketClose }) claimCapability("port", capability) } ``` Once the `setup` function has been called, channels can be created through the IBC routing module between instances of the non-fungible token transfer module on separate chains. This specification defines packet handling semantics only, and defines them in such a fashion that the module itself doesn't need to worry about what connections or channels might or might not exist at any point in time. #### Routing module callbacks ##### Channel lifecycle management Both machines `A` and `B` accept new channels from any module on another machine, if and only if: * The channel being created is unordered. * The version string is `ics721-1`. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanOpenInit( order: ChannelOrder, connectionHops: Identifier[], portIdentifier: Identifier, channelIdentifier: Identifier, counterpartyPortIdentifier: Identifier, counterpartyChannelIdentifier: Identifier, version: string): (version: string, err: Error) { // only unordered channels allowed abortTransactionUnless(order === UNORDERED) // assert that version is "ics721-1" // or relayer passed in empty version abortTransactionUnless(version === "ics721-1" || version === "") return "ics721-1", nil } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanOpenTry( order: ChannelOrder, connectionHops: Identifier[], portIdentifier: Identifier, channelIdentifier: Identifier, counterpartyPortIdentifier: Identifier, counterpartyChannelIdentifier: Identifier, counterpartyVersion: string): (version: string, err: Error) { // only unordered channels allowed abortTransactionUnless(order === UNORDERED) // assert that version is "ics721-1" abortTransactionUnless(counterpartyVersion === "ics721-1") return "ics721-1", nil } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanOpenAck( portIdentifier: Identifier, channelIdentifier: Identifier, counterpartyChannelIdentifier: Identifier, counterpartyVersion: string ) { // port has already been validated // assert that version is "ics721-1" abortTransactionUnless(counterpartyVersion === "ics721-1") // allocate an escrow address channelEscrowAddresses[channelIdentifier] = newAddress() } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanOpenConfirm( portIdentifier: Identifier, channelIdentifier: Identifier ) { // accept channel confirmations, port has already been validated, version has already been validated // allocate an escrow address channelEscrowAddresses[channelIdentifier] = newAddress() } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanCloseInit( portIdentifier: Identifier, channelIdentifier: Identifier ) { // abort and return error to prevent channel closing by user abortTransactionUnless(FALSE) } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onChanCloseConfirm( portIdentifier: Identifier, channelIdentifier: Identifier ) { // no action necessary } ``` ##### Packet relay * When a non-fungible token is sent away from its source, the bridge module escrows the token on the sending chain and mints a corresponding voucher on the receiving chain. * When a non-fungible token is sent back toward its source, the bridge module burns the token on the sending chain and unescrows the corresponding locked token on the receiving chain. * When a packet times out, tokens represented in the packet are either unescrowed or minted back to the sender appropriately -- depending on whether the tokens are being moved away from or back toward their source. * Acknowledgement data is used to handle failures, such as invalid destination accounts. Returning an acknowledgement of failure is preferable to aborting the transaction since it more easily enables the sending chain to take appropriate action based on the nature of the failure. `createOutgoingPacket` must be called by a transaction handler in the module which performs appropriate signature checks, specific to the account owner on the host state machine. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function createOutgoingPacket( classId: string, tokenIds: string[], sender: string, receiver: string, destPort: string, destChannel: string, sourcePort: string, sourceChannel: string, timeoutHeight: Height, timeoutTimestamp: uint64): uint64 { prefix = sourcePort + '/' + sourceChannel // we are source chain if classId is not prefixed with sourcePort and sourceChannel source = classId.slice(0, len(prefix)) !== prefix tokenUris = [] tokenData = [] for (let tokenId in tokenIds) { // ensure that sender is token owner abortTransactionUnless(sender === nft.GetOwner(classId, tokenId)) if source { // we are source chain, escrow token nft.Transfer(classId, tokenId, channelEscrowAddresses[sourceChannel], null) } else { // we are sink chain, burn voucher nft.Burn(classId, tokenId) } token = nft.GetNFT(classId, tokenId) tokenUris.push(token.GetUri()) tokenData.push(token.GetData()) } NonFungibleTokenPacketData data = NonFungibleTokenPacketData{ classId, nft.GetClass(classId).GetUri(), nft.GetClass(classId).GetData(), tokenIds, tokenUris, tokenData, sender, receive } sequence = Handler.sendPacket( getCapability("port"), sourcePort, sourceChannel, timeoutHeight, timeoutTimestamp, protobuf.marshal(data) // protobuf-marshalled bytes of packet data ) return sequence } ``` `onRecvPacket` is called by the routing module when a packet addressed to this module has been received. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onRecvPacket(packet: Packet) { NonFungibleTokenPacketData data = packet.data // construct default acknowledgement of success NonFungibleTokenPacketAcknowledgement ack = NonFungibleTokenPacketAcknowledgement{true, null} err = ProcessReceivedPacketData(data) if (err !== null) { ack = NonFungibleTokenPacketAcknowledgement{false, err.Error()} } return ack } function ProcessReceivedPacketData(data: NonFungibleTokenPacketData) { prefix = data.sourcePort + '/' + data.sourceChannel // we are source chain if classId is prefixed with packet's sourcePort and sourceChannel source = data.classId.slice(0, len(prefix)) === prefix for (var i in data.tokenIds) { if source { // we are source chain, un-escrow token to receiver nft.Transfer(data.classId.slice(len(prefix)), data.tokenIds[i], data.receiver, data.tokenData[i]) } else { // we are sink chain, mint voucher to receiver prefixedClassId = data.destPort + '/' + data.destChannel + '/' + data.classId nft.CreateOrUpdateClass(prefixedClassId, data.classUri, data.classData) nft.Mint(prefixedClassId, data.tokenIds[i], data.tokenUris[i], data.tokenData[i], data.receiver) } } } ``` `onAcknowledgePacket` is called by the routing module when a packet sent by this module has been acknowledged. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onAcknowledgePacket(packet: Packet, acknowledgement: bytes) { // if the transfer failed, refund the tokens if (!acknowledgement.success) refundToken(packet) } ``` `onTimeoutPacket` is called by the routing module when a packet sent by this module has timed out (such that it will not be received on the destination chain). ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onTimeoutPacket(packet: Packet) { // the packet timed-out, so refund the tokens refundToken(packet) } ``` `refundToken` is called by both `onAcknowledgePacket`, on failure, and `onTimeoutPacket`, to refund escrowed token to the original sender. ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function refundToken(packet: Packet) { NonFungibleTokenPacketData data = packet.data prefix = data.sourcePort + '/' + data.sourceChannel // we are the source if the classId is not prefixed with the packet's sourcePort and sourceChannel source = data.classId.slice(0, len(prefix)) !== prefix for (var i in data.tokenIds) { if source { // we are source chain, un-escrow token back to sender nft.Transfer(data.classId, data.tokenIds[i], data.sender, null) } else { // we are sink chain, mint voucher back to sender nft.Mint(data.classId, data.tokenIds[i], data.tokenUris[i], data.tokenData[i], data.sender) } } } ``` ```typescript theme={"theme":{"light":"github-light-high-contrast","dark":"github-dark-high-contrast"}} function onTimeoutPacketClose(packet: Packet) { // can't happen, only unordered channels allowed } ``` #### Reasoning ##### Correctness This implementation preserves token non-fungibility and redeemability. * Non-fungibility: Only one instance of any token is *live* across all the IBC-connected blockchains. * Redeemability: If tokens have been sent to the counterparty chain, they can be redeemed back in the same `classId` & `tokenId` on the source chain. #### Optional addenda * Each chain, locally, could elect to keep a lookup table to use short, user-friendly local `classId`s in state which are translated to and from the longer `classId`s when sending and receiving packets. * Additional restrictions may be imposed on which other machines may be connected to & which channels may be established. ## Further Discussion Extended and complex use cases such as royalties, marketplaces or permissioned transfers can be supported on top of this specification. Solutions could be modules, hooks, [IBC middleware](../ics-030-middleware) and so on. Designing a guideline for this is out of the scope. It is assumed that application logic in host state machines will be responsible for metadata immutability of IBC tokens minted according to this specification. For any IBC token, NFT applications are strongly advised to check upstream blockchains (all the way back to the source) to ensure its metadata has not been modified along the way. If it is decided, sometime in the future, to accommodate NFT metadata mutability over IBC, we will update this specification or create an entirely new specification -- by using advanced DID features perhaps. ## Backwards Compatibility Not applicable. ## Forwards Compatibility This initial standard uses version "ics721-1" in the channel handshake. A future version of this standard could use a different version in the channel handshake, and safely alter the packet data format & packet handler semantics. ## Example Implementations * Implementation of ICS 721 in Go can be found in [bianjieai/nft-transfer repository](https://github.com/bianjieai/nft-transfer). ## History | Date | Description | | ------------ | -------------------------------------------------------- | | Nov 10, 2021 | Initial draft - adapted from ICS 20 spec | | Nov 17, 2021 | Revised to better accommodate smart contracts | | Nov 17, 2021 | Renamed from ICS 21 to ICS 721 | | Nov 18, 2021 | Revised to allow for multiple tokens in one packet | | Feb 10, 2022 | Revised to incorporate feedbacks from IBC team | | Mar 03, 2022 | Revised to make TRY callback consistent with PR#629 | | Mar 11, 2022 | Added example to illustrate the prefix concept | | Mar 30, 2022 | Added NFT module definition and fixed pseudo-code errors | | May 18, 2022 | Added paragraph about NFT metadata mutability | | Nov 08, 2022 | Added `tokenData` to PacketData | | Dec 14, 2022 | Added `classData` and `memo` to PacketData | | Dec 15, 2022 | Tightened spec on `classData` and `tokenData` | ## Copyright All content herein is licensed under [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0).