this is my first ever model I upload. I already gained some experience with finetuning models for SD 1.5 and 2.1. Now I made the step to SD XL. I proudly present:

traditional oil painting 1.0

Goals of this model:

The model is intended to repoduce highly detailed landscape paintings made in the 18-19th century. It is trained on only the finest paintings from the time that fit my own taste. This means the style is almost exclusively a very sharp almost realistic look. I also included some hudson river school painters (marked with *). By the way, this originally was intended tob e an exclusive hudson river shool model.

Technical details:

-Dataset around 365 pictures adding up to more than 8 GB all with a total pixel count greater than 1024x1024

-Training in 14 epochs á 5 repetitions. it had a quite high learning rate of -7*10^7 which is the reason it got destroyed after 14 epochs. I looked at the point where it became more and more bad and that was epoch 7, so there you go.

Limitations:

This is my first model created with SD XL and my first model ever released, so please be gentle with me. Known problems include:

-hudson river shool painters often lead to way to high contrast, sometimes the sky gets pink or strangely colored

-the AI has absolutely no idea where and for what ropes in ships are used, so prompt sailing ships with caution xD

-with „normal“ eg. 1024x1024 resolution the AI has problems filling it up with all the detail -> the picture looks undtailed, I would suggest using hires Fix in automatic1111 and generating 2048x2048. while simultaneously using a little higher base resolution

This is the list of all painters in the dataset:

Albert Edelfelt

Anton Melbye

Claude Lorraine

Jacob van Ruisdael

Johan Dahl

Meinert Hobbema

Paulus Potter

Peter Rubens

Philips Wouwerman

Rembrandt

Rosa Bonheur

William Boguerau

William Turner

Albert Bierstadt*

Asher Brown Durand*

Frederic Edwin Chruch*

Jasper Francis Cropsey*

Thomas Cole*

Important keywords:

highly realistic (for a paticular highly realistic look)

traditional oil painting (phrase used in all captions)

by >name< (self explanatory)

conifers (describing conifer trees like spruce, Pine, Fir)

broad leafed tree (describing all non conifer trees exept for …)

birch (they are easy to distinguish so the AI knows it well)

Maple (for maple trees, may not actually work well)

Mountain scene (for sceneries involving mountains)

Lively river, spray (for wild rivers with white spray)

Rough sea (self explanatory)

Picture frame (negative for visible edges of picture frames)

Damaged/visible canvas (negative for damages due to ageing of the paint like cracking)

FEEDBACK WELCOME. I am very much a beginner.

The Pictures from the dataset are almost exclusively not copyrighted, so I am of the opinion that this model is from a legal point of view and from a ethically point of view save to use in every way. Still please dont try to make money and If you do, please ask me. And if you merge it in any way: please tell me so that I can see what can be done, no need to ask for allowance though.