Foreword

This is the last LoRA model of several attempts to change the behavior of a bottle in an image. And, I am satisfied with this LoRA model so far.

Goal

The goal should be to get images with a laying bottle on a floor, on the ground, on a table, in the sand at the beach or wherever one wants, if one uses this LoRA model. The bottle should be at first glance transparent with a cap made of cork or other material.

What the Model Does

The LoRA model creates a laying transparent glass bottle with a cap.

The color of the bottle as well as the cap can be modified.

Good Prompt

A good working Prompt looks like

laying on its side transparent red bottle on stony ground, background lava spewing volcano, <lora:EmptyGlassBottleConcept:0.75>

First one describes the bottle a little bit and second one describes the background more or less in detail. Then one has a good base concept.

Weights

With a weight of 1.0 one will get images which looks like the ones I prepared for the training.

Good results got I with weights between 0.5 and 0.7.

Training

After a few failures in creating a good LoRA model, I made suitable images and hoped that these would work better than the previous created images.

In the end I used a reduced set of images. In the end there were only 4 images remaining. Using buckets and mirroring I used for the training 32 images in total.

Training iteration were 3500 in total at the end.

I used as image file names

lying down bottle [1...4].jpg

In the case of this LoRA model I did not use a caption text file for the training.

Trigger Word

The Trigger Word is

glassbottle

The LoRa model will work well without the use of the Trigger Word.

Further Reading

In parallel to the LoRA model development I wrote an article about the reason and motivation for this model. This article can be found here https://civarchive.com/articles/7841/pitfalls-of-a-prompt

Final Words

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