This LoRA will stylize your images with an web ink, ink splatter, watercolor quality. It actually has a diverse range of styles depending on the scheduler/sampler combination you use. This ranges anywhere from watercolor to digital painting, sometimes pen and ink, but the look is consistently that of wet ink that's either dripping or has splatter.

Suggested model strength: 0.3-1.0

Suggested samplers/schedulers:

• seeds_3 / beta

• ddim / kl_optimal (or beta)

• dpm_2_ancestral / ddim_uniform

• euler_ancestral / normal (or beta)

• euler / normal

It absolutely provides you with better results than the base Z-Image Turbo model. While Z-Image can generate watercolor-esque images, they tend to be very faded and weak for lack of better words. They don't look right. This LoRA will absolutely fix that and you can really spend a lot of time getting different looks by adjusting the model strength, sampler, and scheduler.

This was an interesting LoRA to train and surprised me. The loss ended up at 0.4001 which I thought may have been too far on the side of being underfit, but it definitely is not. It's a stable LoRA that works at a variety of strengths. You can use this from around 0.3 all the way up to 1.2. Though I would stick to somewhere between 0.75 and 0.9. Depending on the sampler/scheduler you use, you may want to reduce the strength to 0.6. You'll really want to take the time to experiment with this one as it really has a diverse range of style.