Submitted by KingNFT, also found by ZanyBonzy and ArmedGoose
https://github.com/code-423n4/2023-12-revolutionprotocol/blob/d42cc62b873a1b2b44f57310f9d4bbfdd875e8d6/packages/revolution/src/CultureIndex.sol#L209 
https://github.com/code-423n4/2023-12-revolutionprotocol/blob/d42cc62b873a1b2b44f57310f9d4bbfdd875e8d6/packages/revolution/src/VerbsToken.sol#L193
CultureIndex.createPiece() function doesnt sanitize malicious charcacters in metadata.image and metadata.animationUrl,  which would cause VerbsToken.tokenURI() suffering various JSON injection attack vectors.
That is the final metadata used by the NFT (VerbsToken) is not the art piece users vote. This attack could be benefit to attackers, such as creating NFTs containing same art piece data with existing high price NFTs. And this attack could also make the project sufferring legal risks, such as creating NFTs with violence or pornography images.
More reference: https://www.comparitech.com/net-admin/json-injection-guide/
As shown of createPiece() function, there is no check if metadata.image and metadata.animationUrl contain malicious charcacters, such as ", : and ,.
Adverary can exploit this to make VerbsToken.tokenURI() to return various malicious JSON objects to front end APP.
For example, if attacker submit the following metadata:
During voting stage, front end gets image field by CultureIndex.pieces[pieceId].metadata.image, which is ipfs://realMonaLisa. But, after voting complete, art piece is minted to VerbsToken NFT. Now, front end would query VerbsToken.tokenURI(tokenId) to get base64 encoded metadata, which would be:
In the front end, we use JSON.parse() to parse the above data, we get image as ipfs://fakeMonaLisa.

Image link: https://gist.github.com/assets/68863517/d769d7ac-db02-4e3b-94d2-dfaf3752b763
Below is the full coded PoC:
And, test logs:
Sanitize input data according: https://github.com/OWASP/json-sanitizer
rocketman-21 (Revolution) confirmed
0xTheC0der (Judge) commented:
Note: See full discussion here.
