For researchers, art-tech purists, and extreme quality enthusiasts.
This is a non-compromise, zero-quantization LoRA trained on an ultra clean master dataset of 800 curated high-resolution artworks at native 1024x1024.
Architecture: Z-Image Base
Rank / Alpha: 2048 / 2048 (Full 10 GB weights)
Precision: Full Unquantized (No FP8, no compression)
Dataset: 800 Masterworks @ 1024x1024
Key Focus: Captures oil stroke impasto, master craquelure texture, volumetric sea translucency, and Aivazovsky's signature light interplay.
Note for standard GPU users: This 10 GB build is an uncompressed research checkpoint. A pruned/distilled version optimized for everyday generation will be released shortly.