Foreword

In the short time I'm currently learning how to create Embeddings or Textual Inversions and Hypernetworks, there's one thing in particular that I'm focusing my efforts on. I want to be able to create a photorealistic image in the end. When I started creating this hypernetwork, I had in mind that the resulting image should show a mushroom that everyone says looks like a real mushroom.

Hypernetwork vs. Embedding

I prepared an Embedding using in parts the same set of images as I did for this Hypernetwork. As expected, with the Embedding created images, looks like the images, I put into the training process.

In contradiction the Hypernetworks does, what I had primary in mind. I can now create realistic looking mushrooms. Similarity with real existing mushrooms is still rather coincidental.

What the Hypernetwork Does

The Hypernetwork model makes mushroom look like real mushrooms. As stated earlier, a created mushroom looks natural or realistic, but he is not. He is a mix of all the images I used to learn the system how a mushroom can look like.

Test Case

Using a test case one can test how the Hypernetwork influences the created result using a special model.

Simple prompt without using the Hypernetwork:

mushroom

Simple prompt using the Hypernetwork:

mushroom, <hypernet:realmushroom_v10:1>

Normally, there will be a big difference in the look of the created pictures. With Hypernetwork the images will look more natural. Natural means not that there will be a real fungus, which look similar. It is more the look and feel of the mushroom in a more realistic environment.

Training

For the training I prepared a set of free of charge images of mushrooms of different kind. There was still in the center of the image there was most of the time one mushroom in his natural environment like forest or meadow. There is a wide range in the colors of the mushrooms. The image I prepared for training have a resolution of 1024 x 1024 pixel.

Versions

Versions I am preparing can be tested independent from each other. The difference is the number of training steps and the number of images I trained the Hypernetwork on.

Credits

The images for this post were created and/or tested with the following models, among others. My thanks goes to their creators for there good work.

• AbsoluteReality

• DreamShaper

• CyberRealistic

• SDXL Unstable Diffusers

• FUSIONCORE

• Real Dream Classic SD1.5

• Realistic Vision V6.0 B1

• AI Infinity Realistic

Finally

Have a nice day! Have Fun! Be inspired!