GlassSphereLoRA

Foreword

When I started with AI image generation, I adapted the so-called Glass Ball Photography. Unfortunately, my first results were not really reproducible. They were more a product of chance. One major reason was, that I was not yet able to create good Prompts.

Now that I am in a position to create models like LoRAs myself, I am also taking up the aforementioned topic with respect to Glass Ball Photography again.

The first results are very promising. Since I can now create better Prompts, I can see what the result is with a good Prompt with and without LoRA.

Evolution from Version v1.0 to v1.2

A few words are needed about the individual versions. The main difference is the increased number of training iterations from 1000 up to 3000.

• v1.0 → 1000

• v1.1 → 1500

• v1.2 → 3000

The number of training iterations influences the result of the created image considerable.

With increasing weighting, images that resemble the training images are created faster and faster. The influence of the Prompt decreases with increasing the weight. The higher the version number, the faster a glass ball is reached with a given weighting, latest with a weight of 1.0 as value.

Try out and use the model which fits best to your needs.

What the Model Does

The model creates a main glass sphere in the center of an image with or without content and with or without a background, mostly.

Trigger Words

On should use the Trigger Words next to model and weight like this

glass sphere, <lora:GlassSphere:0.5>

The expression or Trigger Words Glass Sphere triggers the model to create a glass sphere. Without the Trigger Words Glass Sphere there will be no result.

Weight

In principle the LoRA model works with weights from 0.0 up to 1.0. Using the weight 0.0 has no influence on the image generation. This weight can still be used for testing purposes. On can start with 0.0 and can go up to 1.0, incrementing by steps of 0.1 to see, how the created result is changing. One has take into account, that the Seed is not changing, while doing this.

Training

The model was trained using six glass spheres of different size and color. The background color of these glass ranged from dark grey over light blue to white.

Conclusion

This LoRA model is the first model that actually does what I had in mind. It is possible to create a glass sphere that either floats in space or lies in the grass, for example. One can put things in the glass sphere or using the glass sphere as itself.

Finally

Have a nice day. Have fun with the model. Be inspired!