Thanks for taking a look at the Ray- series of models.
Back in November-December 2023, I trained the first model in the series, RAYMNANTS, on a series of thousand-ish personal assets (street photo RAW or edited, digital paintings, etc.) with the goal of creating a model that would make people that look like people, and with the visual style I like best (slightly grainy DSLR photo). From there, I trained two additional family of models with different styles that I link below as well.
I kept the models under wraps as my employer was interested in it but since it ended up in a dead-end and with the recent Flux models explosion, I figured I might as well give my work back to the community before SDXL falls into complete obsolescence.
Thank you all for the inspiration, knowledge and keeping the dream alive.

I hope you folks will have a good time with my models~

R.

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Check out my realistic model: RAYMNANTS
Check out my stylized model: RAYBURN
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Introducing RAYCTIFIER1.5

My artistic model, geared towards painting style and all around illustrations, with a classical painting flair.

I’ve finetuned a branch of my realistic model, RAYMNANTS, on about 600 public domain paintings from the Smithsonian online archive, concentrating on the 1750-1850 period.

The result is a model that absorbed some of the style from those classical painting while retaining above average photo capacity. It will perform really well for any prompt destined to output a painting, do well on illustration, but generally underperforms in photo-style context.

HOW TO PROMPT

• Keep it simple:
  Simple media keywords like “Painting of” or “Concept art of” work well for instance.

• Just start with subject, then add more details. The model will react well to painting-related keywords, like sfumato, chiaroscuro, impasto, odalisque, etc.

• Use negative only if you need it, the more generic image correction negative you will use the more the output will start looking generic. Furthering this, RAYCTIFIER functions best without generic embeddings.

Quick tip: Don’t hesitate to layer specific style keywords to really stress out the painting style you’re going for.
Additionally, you can crank up the CFG/lower the steps to simulate different media, i.e. higher CFG/lower steps to have a more acrylic painting style look.

EXAMPLES

1. a 16th century painting of a werewolf howling at the moon, standing on a 19th century London roof at night, black night, black clouds, moon crescent, underexposed, dark, sfumato, chiaroscuro, brown and black tone

2. acrylic painting close up portrait of an old Kenyan woman in pain, face writhing in pain, grey eyes, wide open eyes, looking at the viewer, curly white and grey hairs, wearing a turtleneck sweater, heavy paint, impasto, desaturated

3. a painting of a neo noir detective, wearing a trenchcoat and a hat, walking in the dark streets of Chicago at night, haze, fog, dark, chiaroscuro, sfumato, contrasted

SETTINGS

• Stay within SDXL picture dimensions for the initial gen if you can. (as a rule of thumb adding height and width together should more or less fall within the 2000 range, i.e. 1024+1024=2048 or 832+1216=2048. 1900 to 2100 is okay, 2150 upwards with probably start repeating latent blocks. Multiples of 64 work better too!)

• The model performs well with a variety of sampler, I have a personal preference for the classic DPM++ 2M Karras, 20 steps, CFG 4-6; but some of the newer sampler/schedulers work well too, like HEUNPP2/DDIM Uniform for instance.
  Cranking up the CFG will start burning up the colors, but by lowering the steps as well, you can achieve a pretty decent heavier paint style, for instance CFG12 at 14 steps with the HEUNPP2/DDIM Uniform combo will give you a pretty saturated and convincing heavy gouache style.

• I normally don’t really use upscalers with RAYCTIFIER and usually just perform a second image-to-image step with a resized bitmap at low denoising, e.g. 0.2-0.3 using the same settings in general. You can crank up a bit the denoising and use “detailed face” or something similar as your second pass positive prompt to help with those blurry, painterly faces.

KNOW ISSUES/QUIRKS

• Since it was trained on a lot of classical paintings, the general art style that results in terms of painting will tend to be closer to that. Not using a painted media (painting, illustration, concept art, etc.) will oftentimes give you photo-style results where the model usually will fall flat on it’s face (i.e. it’s not very good at it). If you want photo, I recommend using RAYMNANTS 3.0 instead.

• As a follow-up to the previous, RAYCTIFIER will sometimes be stubborn and ignore styling elements that go against it’s personal taste: flat color, stylized, vector are all examples of things the model will pretend you just didn’t use unless you weight heavily. And then, if you weight it too much, it will just try to piss you off my misinterpreting it. Yes, it’s a moody one.