Instructions to use Omnico/Krea2_turbo_diff_loras with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Omnico/Krea2_turbo_diff_loras with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("krea/Krea-2-Turbo", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("Omnico/Krea2_turbo_diff_loras") prompt = "A photo-realistic photograph of a adult woman with long blonde hair, wearing a brown parka with a fur-lined hood. the woman, who appears to be in her late s or early twenties, has a serious expression and is resting her chin on her hand. she is looking directly at the camera with her blue eyes. her blonde hair is styled in loose waves and falls down her back. she is wearing a beige parka with a fur-lined hood, which is slightly open, revealing her bare shoulders. the parka has a zipper closure and two pockets on the front. the background is blurred, but it appears to be an outdoor setting with trees and a cloudy sky. the image has a sepia tone, giving it a melancholic atmosphere. the overall mood of the image is somber and contemplative.,," image = pipe(prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Best rank to use?
#8
by Anik69420 - opened
Hi, thanks for taking the time and effort to make these. They are great, but I was wondering which rank to use for generation? I am still kinda new to this, so I don't really know much. From your experience, what is the best rank to use?
Hi!
The models that actually modify the base Krea2—and that I genuinely use in my work—have a rank of 128. I believe a rank of 256 is excessive (though I haven't extensively tested every model in this regard). If you are short on storage space or VRAM, you could try a rank of 64 or even 32. If you only need to capture the style, a rank of 8 or 16 might suffice.