How to use publicprompts.art dreambooth models with InvokeAI
Extending the dream
Public Prompts has some really good extra models trained on some fun internet things. Last night I figured out how to get them loaded into my InvokeAI Stable Diffusion install.
Here is how you load it up:
Download a model:
We will be using the Pixel Art model, but the same setup works for
App Icons. Download the model from the page below.
Pixel Art V1 (DreamBooth model):
Link model [Linux and MacOS]
We get symbolic links in Mac and Linux-land. I am not sure what goes on in the Windows world, so you will have to search for info on that. I keep my models in a directory away from the Stable Diffusion code base so that I can rebuild at will and not have to download gigs of data again.
Create a directory to hold the PublicPrompts models:
I put mine in the
Then create a symbolic like this:
ln -s /Users/$HOME/bin/models/Pixel_Art_V1_PublicPrompts.ckpt /Users/$HOME/bin/InvokeAI/models/ldm/PublicPrompts/Pixel_Art_V1_PublicPrompts.ckpt
Edit the Models config
With the new version of InvokeAI they made it easier to load up different models.
You just need to configure them to point it at our symbolic link from above.
Then we name it whatever you'd like to call it in the args.
configs/models.yaml with your favorite editor. Then add the following.
Note: YAML syntax is specific, so copy what they already have in place.
pixel_art: config: configs/stable-diffusion/v1-inference.yaml weights: models/ldm/PublicPrompts/Pixel_Art_V1_PublicPrompts.ckpt width: 512 height: 512
Start InvokeAI with your new model
Now that we have the model setup, all we need to do is start up the web UI with the new model.
python scripts/invoke.py --web --model pixel_art
During loading, you should see it load your new model:
>> Loading model from models/ldm/PublicPrompts/Pixel_Art_V1_PublicPrompts.ckpt
Each new model has a bit of a trigger phrase to use. Add
, in SKSKS art style to the end to trigger the model.
Some suggested configurations:
Suggested CFG scale: 10-15 [I used 12]
Suggested steps: 20
dog, in SKSKS art style
cat, in SKSKS art style
monkey, in SKSKS art style
mushroom, in SKSKS art style
Mario didn't work out, so you get a mushroom.
astronaut, in SKSKS art style
Now that you know how to add extra models to your InvokeAI install you can
App Icons and
Microworlds models from publicprompts.art
Extra Extra Credit
With this new found knowledge you could get the Dreambooth models from
Hugging Face and drop them in to your install (I haven't tried, yet.).
If you have any questions or comments, please feel free to reach out to me on Mastodon.
The featured image was generate with the prompt:
skate or die, in SKSKS art style. I have no idea where it came from, but I like it.