Training a Model at Deepmode

Hey Deepmoders,  

I want to share a few tips and tricks from our community on getting the best results when training your model at Deepmode.

If you want to learn more about model training keep reading or watch this video:

We have covered training a new model a few times here and here, but based on my testing and feedback from the Deepmode Discord server I have a few more tips for getting the best results.

Quality images 

The most important thing to focus on is your overall image quality, make sure all of your images are upscaled and as high res as possible. This allows for finer details to be seen and offers the AI a clearer look at facial features.

Amount of images 

Deepmode allows for up to 30 images for training purposes but you don't need 30 for a quality result. In fact, some users, including myself, have found some great results from 16 images, as follows:
 
10 portrait images showing the shoulders up. (Remember to include several different angles) 

4 medium shots showing the waist up. Some users have recommended using 2 clothed images and 2 naked images to get a consistent body shape 

2-4 full-body shots. Again, mix in some naked shots here if you are looking for a particular base body shape. 

Be sure to edit your photos so that there are as few defects or distractions as possible. In my experience, this includes the likes of facial piercings and heavy makeup. My original version of Farah would often be generated with nose piercings and heavy contour makeup because a couple of my original references had these features. 

Try to include several different angles of the face including straight-on, semi-profile, and side profile.

It's worth noting that a variety, in terms of hairstyles, clothing, and background, will help the AI create a wider range of styles for your model. If all of your images feature the same hairstyle, clothing, and background it may be more difficult to produce images featuring variety. Just make sure all of your images are focused on your model and don’t have any other people, or at least too many distractions, in the background.

Following these tips should help you get better generations from your models. If you have found any other methods that help you out, let us know over on our Discord server. 

Thanks

Farah

Train your virtual model now at Deepmode.ai

Check out a video version of this tutorial on my X page

Daily News for Curious Minds

“I stopped watching the news, so sick of the bias. Was searching for an alternative that would just tell me WHAT happened, with NO editorializing. I found it. It’s called 1440. It assumes you are smart enough to form your own opinions.”

Sign up now!