I tried to remove the old yellow tint for V0.2, changed the captions a bit, and added some images to the dataset.

The main drawback of V0.2 is that signatures appear more often at the bottom of the images.

This checkpoint is a fine-tune from of the base SD XL model.

The objective of this model is to reproduce traditional painting styles and artistic representations of classical themes.

The trained artists are :

EBls = Eugen von Blaas

Bnhr = Rosa Bonheur

bchr = François Boucher

WllBgr = William Bouguereau

Cbnl = Alexandre Cabanel

Crvgg = Caravaggio

Chpln = Charles Chaplin

edgs = Edgar Degas

DLCRX = Eugène Delacroix

dlflt = Albert Edelfelt

DlpEjl = Delphin Enjolras

LRFlr = Luis Ricardo Falero

frnt = Émile Friant

grm = Jean-Léon Gérôme

gdwrd = John William Godward

ltr = Henri Fantin-Latour

EBLght = Edmund Blair Leighton

AlfMch = Alphonse Mucha

rmnd = Raimundo de Madrazo y Garreta

TdRll = Theodoros Ralli

rmbrndt = Rembrandt van Rijn

Srgnt = John Singer Sargent

Sgnc = Guillaume Seignac

srll = Joaquín Sorolla

AZrn = Anders Zorn

Since I trained the model with a low learning rate, using the artist tag may not have a huge impact on the image. Employing them with their usual theme can work better.

For resolution, you can use 768x1024, 768x1152, 896x1024, 1024x768, 1024x896, 1152x768.

You can use it with my artist lora for XL to reinforce a specific style.

I plan to update it later by training with a bigger dataset, curently it uses around 5000 images.