Uploading Civitai Models

Civitai is a platform for image generation models. It contains thousands of finetuned models stored on its platform. Through our Civitai integration, you can simply type in the url to any Civitai checkpoint and use it on Flush’s GPUs.

In the example below, we are going to upload and run the popular CyberRealistic Model on Flush using the WebUI. First, we must add CyberRealistic as a new model via its Civitai link.

This model will then appear in My Models. If we click on “View Details”, we’ll see the following:

This page shows how we can deploy Flush. We simply run the commands on the right.

DreamBooth Finetuning

Dreambooth is a method to finetune and personalize stable diffusion models given some images of the subject.

In the example below, we’ll show how to finetune Dreambooth to generate professional headshots of the Rock. First, let’s find four images of the Rock:

It’s important to make sure the images are focused on the subject’s face, and that they are 512x512. If the images are not 512x512, we crop them to this size but we cannot guarantee that they will be exactly focused on the subject completely. Now, let’s create a finetuning run.

This model will then show up in My Models. If we click on “View Details”, we’ll see the same as above. If we want to call the model, we must refer to the subject type as “sks ”. Therefore, in our case, our subject is “man”, so we refer to it as “sks man”.

As such, if we want to generate a professional headshot, we can use the following prompt:

"a photo of sks man, 8k, linkedin headshot, close up, high quality, upper body,"

If we pass this as the prompt in playground (or using our SDK), we’ll get something like this: