Dear friends,

it’s nice to be back in Berlin and to continue my AI experiments.

Today I want to talk a bit about StableDiffusion models. (I keep it simple)

All AI tools like ChatGPT, Midjourney or StableDiffusion use models. A model is like a shoebox with a lot of data inside. The AI tool uses this data to generate a text, an image or a video.

StableDiffusion is an open-source software to produce images and videos with AI. Depending on the model you connect it with, it generates images.

If the model contains only images of dogs, StableDiffusion will only generate dogs. If the model only contains underwater pictures, it can only produce images which look like they were taken under water.

There are general models called Base Models. They contain images of everything. They include images of people, houses, cars, trees, you name it. This is why they are so big. Sometimes several Gigabytes. The Juggernaut XL is a very good one.

With those Base Models it is possible to generate almost everything.

Then there are more specified, smaller models called Loras which you can connect additionally. Those contain images of a specific style or specific objects or people.

There could be one Lora (Small model) which only contains images of snow. So when activated, all you images will be covered with snow.
Have a look here: Aether Snow.

With Stable Diffusion it it possible to combine different Loras and even adjust their strength.

One thing to mention is that all models I will feature here are SDXL models. This means they are made to create images in a high resolution like 1024x1024px.

Online you will find a lot of models which are older and not optimised for high resolutions. They are called 1.5 models. They are nice as well but only generate images with a resolution of 512x512px.

So, here is a list of some of my favourite SDXL models and Loras which you can use with StableDiffusion:

Base Models

General models which can generate all kind of images. They are used as a base.

• Juggernaut XL (My favourite base model. I use it almost all the time. Great for generating realistic people)
• Dreamshaper XL (A good alround model, a bit too fantasy for my taste. But good to test and play around with)
• SD XL (This is the default Stable Diffusion XL model. It generates the most neutral images and does not really have an own style.)
• SDVN7 NijiStyleXL (For everybody who likes Anime, this is a great base model for it)
• Anime Art Diffusion XL (Another great Anime model)

Loras

Loras are small models which contain images of a specific style or topic. They need to be combined with a base model from above.

• Detail Tweaker XL (A very cool Lora which adds more details to your image. I recommend that a lot. You can always play around with the strength)
• Pixel Art XL (Turns everything into retro pixel-art)
• Lelo Lego XL (Turns everything into Lego)
• Satoshi Urushihara (For that late 80’s early 90’s anime aesthetic)
• Crayon Style SDXL (Everything painted by a child with crayon. With saturated colours, this can produce beautiful images)
• Samaritan 3D Cartoon (Turns everything into a 3D cartoon)
• Blacklight Makeup (Switches on the blacklight and lets things glow in neon)
• Fx Monsters (Great, terrifying monsters)
• Fire Element Special Effects Yuan (Let it burn)
• SDXL Ms Paint Portraits (One of my favourites. Turns everything into a crappy Microsoft paint drawing)
• PS1 Graphics (Turns everything into the look of a Playstation 1 game)
• Split (Splits a person in half and shows their skeleton between)

I am also working on a new platform with the name Visual AI Mastery.

There I will publish much more information about AI tools for creatives. So who wants to dig deep into this — keep posted.

Thank you for reading!

Much love,
Marius

Thanks for reading Marius Jopen! Subscribe for free to receive new posts and support my work.