AI art NFT metadata builder
Minting AI artwork as an NFT needs a metadata JSON file that marketplaces
read to display the piece. The de-facto standard is the ERC-721 metadata
schema — name, description, image, and an attributes array. This
builder collects your fields and AI generation details and outputs a clean,
compliant file ready to pin to IPFS.
How it works
You fill in the core fields and any generation attributes — model, seed, style
and a prompt hash. The tool assembles them into a valid ERC-721 object, encoding
each attribute as a { trait_type, value } pair so OpenSea and similar
marketplaces can build filters and rarity tables. The seed is marked as a
numeric display type. The output is pure, parseable JSON with no trailing
commas or invalid fields.
Understanding the ERC-721 metadata schema
The ERC-721 standard defines what a token is but says nothing about what it looks like. The community converged on a metadata schema that extends the contract:
{
"name": "My AI Artwork #1",
"description": "A description of the piece.",
"image": "ipfs://Qm...",
"attributes": [
{ "trait_type": "Model", "value": "SDXL" },
{ "trait_type": "Seed", "value": 42, "display_type": "number" },
{ "trait_type": "Style", "value": "photorealistic" }
]
}
The image field holds a URI to the artwork itself — ideally an ipfs:// content-addressed URI rather than an HTTPS link, so the image remains accessible independently of any single server. The attributes array is where AI generation details belong; marketplaces read these to build trait filters and rarity distributions.
Why AI-specific attributes matter
Recording the model, seed, and style in the metadata does several things for AI artwork specifically:
- Provenance. A collector or auditor can verify which generative model produced the piece, which is increasingly relevant as platforms develop policies around AI-generated content.
- Reproducibility. A documented seed means the image can be regenerated from the same model to confirm authenticity, at least while that model version remains available.
- Rarity mechanics. For a generative collection, traits like model, sampler, or style can drive rarity tables the same way hand-drawn trait layers do.
- Prompt privacy. Storing a cryptographic hash of the prompt lets you prove the exact prompt later (by revealing it and showing it matches the hash) without exposing the recipe to competing creators while the work is current.
Tips and notes
- Pin to IPFS for permanence. Use an
ipfs://URI for the image so the art outlives any single server. - Store a prompt hash, reveal later. Keep the exact prompt private while still being able to prove provenance.
- Use numeric display types for the seed and any edition number so marketplaces sort them correctly.
- Validate before minting. Paste the output into a JSON validator — a single malformed character can make a marketplace fail to render your token.
- Save the raw prompt separately. Metadata is immutable once minted; keep the original prompt and seed in a secure record off-chain so you can reconstruct or prove it later.