Style Buff

💚 for Updates - Is this test lora good, bad? I would love to see your images and reviews!

This is a test LoRa trained via Pixai using 100 upscaled images. I wasn't sure how the dataset might translate to outputs and what sort of styling would occur. So far I've had limited testing with this model and will most likely update things accordingly later on. This is also my first attempt at training anything, using a pre-determined tool preset options at that.

For now, this style seems to enhance images (depending factors, taste etc.) to a certain degree with it's own styling and changes:

• alternate styling

• creativity

• special effects

• lines

• lighting

• eyes

• glow

• etc.

This LoRa might also help shape up older or messy models or models that may not do creative aspects as well. I've noticed some prompts/models/etc. perform better without the LoRa at times depending on seed/sampler/etc. This is a test model afterall and is not intended to be an all-around enhancer all the time!

I'll try to add more image examples and information later on. Most of the dataset images were portrait style images - though this may come through in some cases, it doesn't seem to be limited to portraits only.

Trigger word you can use is stylebuff, feel free to try with and without!

I've noticed ~0.1-1 weights can do fine on the models I've tested, some can take higher values more than others. You might see good results with even higher weights as well. I've yet to try out negative weights.

Note: this is very model/prompt/sampler dependent, use what you find works with your configurations!

If you merge models with this resource, please note what was used and how in your recipes!