Foreword
I created a LoRA model for creating panama hats. I wanted to check what the influence is on a photo when I use such a model in the Prompt. The answer seems to be a little bit complicated. And, the answer followed, what I am seeing creating images and using models.
Prerequisite
That the following will work we have to use the same Seed. Model stays also the same. The weight in the trigger word for the LoRA is still changing.
Used Prompt
For my last study I used the following base Prompt:
masterpiece, intricate photo, beautiful rose wearing panama hat, highly detailed, sharp focus, cozy outdoor lighting, high resolution, 8K, <lora:PanamaHatConcept:0.0>Based on 8 photos of panama hats with different view angles I created a LoRA model for testing purposes.
With a weight of 0.0 in the Prompt there is no effect on the created image. Then the photo is changing. The hat looks step by step more like a panama hat. In the last step it is a panama hat. In the following section I am presenting the created results.
Girl with Panama Hat using a LoRA model
We are starting with this Prompt:
Same Prompt with <lora:PanamaHatConcept:0.0>
Then we using this Prompt:
Same Prompt with <lora:PanamaHatConcept:0.2>
Then we using this Prompt:
Prompt:Same Prompt with <lora:PanamaHatConcept:0.4>
Last but not least we using this Prompt:
Prompt with <lora:PanamaHatConcept:0.6>
The last result is quite funny, but this is, what LoRa does. It is moving the Photo to what we told LoRa it should be. In this case at the end a panama hat.
Simplified Prompt
With a simplified Prompt using the trigger word and a weight of 1.0
panama hat, <lora:PanamaHatConcept:1.0>we get what we trained on, a panama hat:
Conclusion
Based on my experience with the various models, I expected this kind of behavior. I had not yet seen it in this clarity. This seems to be one advantage to using LoRA models.
Finally
Have a nice day. Have fun. Be inspired!