Concepts is the general term the AI hub “Huggingface” uses for different kinds of AI models. Graydient’s vision is to support more AI concepts beyond images, long term, such as audio and video, in one convenient subscription package. In the short term, we’re surely confusing you. We get that.
To make matters worse, People that don’t live/breathe AI people refer to models simply as “art styles” and think that “models” means people. So we have some clean-up to do. Everything is going to be renamed to “recipes” by this summer. Until then, remember that in our software:
Concepts = AI Model, Checkpoint, CKPT, Diffusers that someone created, huge gigabyte files that we download and install for the community at large
Styles = Text. Prompt templates, basically macros for your favorite keywords. These are sometimes private. However, the community has made a bunch of public ones that you can copy and install easily.
Use /render or /remix with <concept> names like this:
DECPRECATION NOTE: THE MODELS COMMAND
The /models command currently pertains to older Dreambooth models created by us in our very early days. These are being phased out.
In the tutorial below, we will try each prompt and find out if it’s true!
When joining a channel that’s already using Pirate Diffusion, start by seeing what /concepts are already defined
/concepts
Example output:
Available concepts: <openjourney>, <urpm13>, <hassan15>, <av3>, <hassaku>, <22h>
Usage:
/render a wizard eating a donut <openjourney>
Result:
Let’s compare that to a model that’s trained on anime, Anything-v3.0, which we renamed to simply <av3> so it’s less typing:
/rendera wizard eating a donut <av3>
We add or update new concepts every week.
Using concepts + styles together
Consider this default Stable Diffusion prompt without a <concept>
/render three japanese women eating udon
Compared to an AI model that was specifically trained on anime.
/render three japanese women eating udon <av3>
This is alright, but if we invoke the name of our favorite anime artists or other high quality words, the results are going to be better. Especially if we invoke positive prompts and negative prompts. See?
/render three japanese women eating udon, high quality, masterpiece, makoto shinkai, [[jpg, low resolution, low quality]] <av3>
You don’t want to type all of that every time.
Let’s make a style called “jp” for anime, like this:
/style /new:jp $prompt high quality, masterpiece, makoto shinkai, [[jpg, low resolution, low quality]]
$prompt refers to the placeholder where your future prompt will go, like making a little program that will type all of these things for you when you call your style.
Put it all together:
/render /style:jp three women eating udon <av3>
Set it and forget it. We can use this anytime in other prompts
/render /style:jp a mecha godzilla eating a waffle <av3>
So short, so much power! What a joy!
Styles are intended to be personal and secret. Some styles are shared in groups. You can use the copy command to move a style into your private bot.