AI Digital Asset Provenance & Perceptual Hash Drift Workbench

Generative Parameters AINFT-GEN-086

Presets

Generative Prompt Text

Random Seed 884920412

Model Pipeline Version

Style Guidance Weight 7.5

Epoch / Iteration Stage 1

Latent Noise Mutation (%) 0%

          dHash (Difference 8x8)
          a3f089e1b2c3d4e5
        

Average Hash (aHash) ff80c0e0f0f8fcfe

          Hamming Bit Distance
          0 bits
        

Prompt Drift (Levenshtein) 0 chars

          Merkle State Root
          0x7d8f...5e2b
        

Procedural Asset Canvas 512x512 Render

8x8 Perceptual Gradient Matrix Cyan: 1 | Dark: 0 | Red: Drift Diff

State commitment hierarchy constructed from SHA-256 leaves of (Generative Prompt, Seed, Model Pipeline, and 8x8 dHash).

Simulated On-Chain State Verification:

Merkle Root 0x7d8f9a2b1c4e6f8a3d5e2b1c4e6f8a3d5e2b1c4e6f8a3d5e2b1c4e6f8a3d5e2b is cryptographically VERIFIED against local leaf state.

Live Dynamic ERC-721/1155 Schema Compliance Schema Valid (100%)

State Mutation Event Log

VERIFIED PROVENANCE ANALYSIS COMPLETE

Status: VERIFIED_GENUINE

Analyzed Asset AINFT-GEN-086 | Current dHash: a3f089e1b2c3d4e5 | Hamming Drift: 0 bits | Merkle Leaf Status: VALID

Procedural assets, compact hashes and local commitments

Your browser does not support video.

Read the explanation

This workbench draws shapes locally. It does not call the model named in its dropdown. A sine-based generator uses the seed to place circles and squares. Style changes each shape radius by one plus style divided by ten. At seven point five, the multiplier is one point seven five; at twelve, it is two point two. The bars share two hundred pixels per multiplier unit. Prompt text, epoch and model name are not inputs to the drawing function. Noise changes hue and adds unseeded random pixel perturbations. A fixed seed alone therefore does not reproduce a noisy drawing. These are procedural canvas images, not generated NFTs or verified model outputs. The drawing is sampled onto a nine by eight grid and converted to rounded luminance. The difference hash compares neighboring values, eight comparisons across each of eight rows, producing sixty-four bits. The average hash uses the first eight columns, sixty-four samples, and compares each with their mean. Both bars show sixty-four output bits at seven pixels per bit. A Hamming distance counts differing hash bits; it does not measure authorship, ownership or semantic identity. There is a subtle baseline rule: matching baseline seed and prompt with zero noise replaces the reference hash, even if style has changed. A resulting zero drift can therefore reflect an updated reference rather than an unchanged original drawing. The local commitment starts with four SHA-two-fifty-six leaves: prompt, seed plus model name, difference hash, and epoch plus style. Two branch hashes combine the hexadecimal leaf strings; one root combines the branch strings. Four leaf hashes plus two branches plus one root gives seven computations, shown on a shared sixty-five-pixel scale. This binds those local strings through this particular encoding. It does not authenticate the claimed model, publish to a blockchain or verify the placeholder IPFS image. Noise is not directly included; it enters only through the compact difference hash. Export includes state, baseline, displayed metrics, audit text and metadata. Its displayed root is abbreviated, while metadata holds the full root. Genuine and proven labels are local UI classifications, not independent provenance proof.