STUDIODIFUSIONBrowse catalog
LoRA

Custom LoRA

A style or identity adapter trained on your material and tuned until it actually holds.

Send 30–300 reference images. We curate, caption, train and evaluate against a held-out set, then iterate until identity or style holds at usable weights. You get the safetensors and the training card.

input
30–300 images
base
SDXL or FLUX.2
output
safetensors + card
revisions
until eval passes
$7997–10 working days

Pay now and a brief form arrives by email so we can start.

Licence · bespoke
Rights negotiated per project and confirmed in writing before work starts.