Why did Azuki's Elementals experience image collision?

Author: Jaleel, Jack

Editor: Jack

As the NFT market enters a bear market, even the market leader "Bored Ape" has not been able to escape, in sharp contrast, Azuki has performed exceptionally well. Over the past three months, Azuki has gone against the market trend and successfully risen to become the strongest blue-chip project. During the peak of the bull market, Azuki's floor price was over 30 ETH, and even a few days ago it rose to 17 ETH. Not only that, but the community atmosphere of Azuki is also rare, with high activity and cohesion even in a bear market.

Until the release of Elements at midnight tonight, the community was on high alert. The short ten-minute window for the whitelist Mint caused the website to crash due to overwhelming traffic in a short period of time. After the sale, the project team transferred 20,000 ETH directly. The most discussed topic in the community is the high degree of feature overlap between Elements and the first generation Azuki, which may directly dilute the value of the first generation Azuki.

At the beginning, the community questioned the high degree of overlap between Elements and Azuki. However, as more and more Elements were revealed, many holders found that their Elements looked exactly the same as those of other holders, not just highly overlapping, but completely identical.

As of now, Elements has been found to have 5 sets of completely identical images. These five sets are numbered: #2210 and #10744; #1077 and #8600; #16046 and #8914; #16580 and #5613; #19697 and #2475. Not only are the front-end images displayed by Elements' NFTs identical, but even the names of the images saved are the same.

As more and more collision situations were discovered during the opening of Elementals, the team realized the problem and responded. Azuki co-founder 2PMFLOW replied to the community on social media, stating that this situation was due to a technical malfunction, where the event logs of the data provider became outdated due to the Ethereum block reorganization, resulting in incorrect processing of metadata for a few tokens. The team is working hard to restore the correct images and metadata.

"Block reorganization" occurs when blocks are generated simultaneously on a blockchain, and errors or malicious attacks can cause temporary duplicate blockchains. Because the miner adding to the next block must decide which side of the fork is the correct or canonical chain, "block reorganization" can occur. Once the miner chooses the fork or canonical chain, the other chain is lost.

Looking back at the last major "blockchain reorganization" incident, it occurred on May 25, 2023. The Ethereum Beacon Chain experienced a reorganization of seven blocks, exposing a high-level security risk known as chain reorganization. Validators on the Eth2 (the current consensus layer upgrade) Beacon Chain became unsynchronized after upgrading to a specific client update.

"撞图"其实并不罕见

"Coincidental images" are actually not uncommon

Regarding the response from 2PMFLOW, co-founder of Azuki, stating that the "collision situation was due to Ethereum block reorganization and is a technical malfunction," scriptmoney.lens, founder of cryptochasersco (@scriptdotmoney), disagrees. He expressed his own opinion on social media that the reason for the collision situation was due to the server not handling exceptions properly while listening to events, not the data provider. In other words, it seems that Azuki's statement that the collision situation was due to a technical malfunction is inconsistent.

When discussing the technical principles of NFT image generation, we need to focus on three core elements: layer styles, generation algorithms, and the upload format of the generated NFT images. Due to the variability of these three factors, countless forms of NFT can be produced.

First of all, layers can be viewed as pre-made materials, such as 10 different colored backgrounds, 10 different hairstyles, 10 different hats, 10 different handheld parts, etc., each of which is an independent layer. Once the number and style of layers are determined, we can preset various influencing factors such as rarity index and target number of generated images to create a set of NFT algorithms that can generate various combinations. Then, according to the generation algorithm, each material layer is stacked from the bottom to the top to form a complete NFT.

According to the number of prepared layers and the complexity of the algorithm by the project party, the probability of collision of the generated NFTs is not the same. For example, if a NFT project party only prepares 3 layers, each layer has only more than 10 kinds of elements, and the generation algorithm is relatively simple, then the probability of collision of the NFT series will be very high. However, for blue-chip projects like Azuki with multiple layers, hundreds of elements, and extremely complex generation algorithms, even if tens of thousands of new NFTs are generated, it is difficult to encounter a collision situation.

Of course, because the process of generating images is completed off-chain, even if there is a collision phenomenon, it can be screened again before the image is officially put on the chain to filter out duplicate images. This requires the project party to fully screen the generated NFT images before putting them on the chain. Even if the project party directly applies the original materials to generate them, there should be a reasonable screening process to ensure the performance of the program and that all combinations strictly follow the preset probability distribution.

So, in summary, what is the problem with Elementals?

First and foremost, it is highly likely that Elements directly utilized the layer, element, and even generation algorithms of Azuki or BeanZ. However, perhaps due to the limited "elemental concepts" of water, fire, electricity, and the like that the NFT series revolves around, the related generation algorithms were not properly debugged, resulting in instances of repeated NFT generation.

Secondly, the community's claim that "minting identical images in the same block" does not seem to be entirely true. According to an NFT technician, the NFTs with duplicate images in the Elementals series do not have sequential or similar numbering, indicating that the probability of them being minted in the same block is small. Therefore, there is another explanation that the Azuki team did not conduct a secondary screening of the generated Elementals, resulting in duplicate images being directly uploaded to the chain.

Of course, no matter what the reason is, it is unacceptable for blue-chip NFT project teams like Azuki to have image collision incidents. Obviously, the team did not do enough preparation before the release of Elementals.

Under the nose change picture, why can the pictures on the chain still be modified?

Regarding the issue of casting and mapping for Elementals, 2PMFLOW, co-founder of Azuki, responded to the community on social media that the team is working hard to restore the correct images and metadata for Elementals mapping, in response to the appearance of identical images in Elementals mapping. Subsequently, BlockBeats also found that when querying the Elementals series contract, some NFT's tokenURL could not display the query results. At the same time, some preview images of Elementals NFTs on Blur and other NFT trading platforms cannot be displayed properly. Azuki team conducted a real-time "image replacement live broadcast" in front of the community.

Already uploaded to the chain and minted NFTs, why can they still be changed and modified? This requires a simple review of the storage method of NFT images.

We know that today's Internet is built on the HTTP protocol, which is a transmission protocol used to transfer hypertext from a network server to a local browser. When we obtain web pages or files through HTTP, we request and retrieve information from the server. This is a centralized process that has multiple issues, such as the inability to access files if the server fails, and the potential for significant pressure on server bandwidth if a large number of users attempt to access the same file simultaneously.

Compared to that, IPFS adopts a more distributed approach. IPFS is a protocol and network designed to create persistent and distributed storage and sharing of files. It is a new hypermedia transfer protocol that addresses a large number of data storage and bandwidth issues. IPFS stores files and other data as blocks on multiple nodes, and uses a unique hash value to identify each block. When a user requests a file, IPFS retrieves the block from the nearest node instead of the original server. This makes data access faster and more stable, and prevents data from becoming inaccessible due to a single server failure.

Because of these advantages, IPFS has become the ideal choice and consensus for storing NFT metadata. The information contained in NFTs usually includes descriptions of art works, such as their authors and creation dates, which are called metadata. Due to the immutability of NFTs, metadata needs to be stored permanently, and the distributed nature of IPFS can provide such a service. At the same time, the distributed nature of IPFS can also prevent the metadata of NFTs from being tampered with, ensuring the security of NFTs.

In the IPFS file system, each file will generate a hash value based on its content, and the file in IPFS will be indexed based on this hash value. It also checks in advance whether this hash value has been stored. If it has been stored, it will be read directly from other nodes without the need for duplicate storage, which in a sense saves space. This greatly improves the efficiency of resource storage and reduces transaction costs.

After querying the Elementals contract, we found that the image metadata of Elementals was not uploaded to IPFS, but instead stored on Azuki's centralized server. Therefore, ownership and modification rights of the metadata are held by the Azuki team and can be changed at any time.

Actually, it's not just the Elementals project that chooses to store images on their own server. Many "wealthy" blue-chip NFTs do the same to gain greater control over their projects. However, compared to Elementals, the metadata of the previous Azuki series was stored on IPFS. So why did the team "reverse" the decentralization process of the project? Perhaps everyone realized from the beginning that Elementals might have problems.

And even more outrageous, during the metadata modification process, the team made a new mistake. scriptmoney.lens found that the fix for metadata uploaded by the Azuki team seems to be a file, so it jumps to the download page as soon as it is accessed. In other words, the server did not upload the normal json metadata code, but directly uploaded image files. Therefore, when third-party trading platforms like Blur search for NFT image data on the server, they cannot read it properly, which is the problem presented at the beginning of this paragraph.

When we look at the downloaded files pointed to by the metadata, we find that these new images cannot even be described as "repaired". Taking Elementals #16580 as an example, as of the time of writing, there is not a single identical element between the metadata file of this NFT and the original version (see the figure below). Although the team had previously announced their image replacement action, such a large difference may be difficult for many people to accept.

No matter how you evaluate it, the release of Elementals is a catastrophic event. It not only hits the Azuki community itself, but also the entire avatar NFT field. From experience, people's confidence and trust in NFT project parties are often very fragile, and Azuki's vampire behavior is rapidly consuming the only remaining faith in this industry.

Of course, in the face of community questioning and criticism, the Azuki team has also responded positively, stating that they will work hard to make up for their mistakes and announce a new "Green Bean" airdrop this afternoon. Although the scene of Elementals is very chaotic, perhaps we can still leave some faith and give the Azuki team one last chance?

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia