Slider controlling background focus.
Dataset is composed of 100 images, 50% photo, 50% illustrations.
Usage:
- Positive: sharper, more detailed backgrounds.
- Negative: softer, out-of-focus backgrounds.
"Limitations":
Due to the way it is trained (predicting the difference between 3 different captions) it will change the image a bit.
Higher/Lower values will shift the composition a little. The model may add extra background detail such as people or signage, or zoom in on the subject.
Might sharpen and blur the subject itself a bit depending on the slider side.