---
tags:
- text-to-image
- lora
- diffusers
- template:diffusion-lora
widget:
- output:
    url: samples/style_transfer_samples.png
  text: 'Style transfer samples'
- output:
    url: samples/Count_BinFace_samples.png
  text: 'Count BinFace samples'
- output:
    url: samples/blue_eye_samurai_samples.png
  text: 'Blue Eye Samurai samples'
- output:
    url: samples/cyanide_happiness_samples.png
  text: 'Cyanide & Happiness samples'
- output:
    url: samples/blueprint_wireframe_samples.png
  text: 'Blueprint / Wireframe samples'
base_model: krea/Krea-2-Raw
instance_prompt: null
license: apache-2.0
---

# Krea 2 Premium LORAs

<Gallery />

## Model description 

Woah, woah, woah, *PREMIUM LORAs* ? What's this all about ? I'm glad you asked, since the company that shall not be named decided to call 
their [100 step abominations](https://huggingface.co/ilkerzgi/fal-Krea-2-Style-LoRAs) "style" LORAs, I had to adapt and somehow distance myself from that, so here we are:
Dever's Krea 2 Turbo Premium Style LORAs that were not trained in under 2 minutes to spam HuggingFace. Rant over.

#### Style Transfer RC1 (experimental)

It was trained to keep the composition of the first reference image and apply the style of the second reference.
Use `KSTRANSFER` as the trigger. [See the workflow example from here](https://huggingface.co/DeverStyle/Krea-2-Premium-Loras/tree/main/workflows)
Uses [Krea2 edit nodes from Ostris](https://github.com/ostris/ComfyUI-Krea2-Ostris-Edit)


#### Count BinFace character

Use `Count BinFace` as the trigger. [Who is Count BinFace ?](https://en.wikipedia.org/wiki/Count_Binface)

#### Blue Eye Samurai style

Use `dvr_bes_style, ... your prompt ...` 

#### Cyanide and Happiness style

Use `dvr_ch_style, ... your prompt ...` as the trigger word, RECOMMENDED to add ` stick figures ` to force the stick figures specific to the show. LORA samples show the same prompt and seed with the LORA off and on. 
The dataset has a bias towards characters so prompts without any people might not be as great. Similarly very long prompt with a high emphasis on realism might not work as well either.


#### Blueprint / Wireframe style

Use `dvr_wf_style, ... your prompt ...` as the trigger word. LORA samples show the same prompt and seed with the LORA off and on.
Prompts used for samples are from [Krea's prompting guide](https://github.com/krea-ai/krea-2/blob/main/docs/prompting.md)


### Download models

[Download](/DeverStyle/Krea-2-Premium-Loras/tree/main) them in the Files & versions tab.

#### Disclaimer and Important notes

The LoRAs in this repository are created for research purposes only and are not intended for commercial use. Some might be based on TV series which are protected by copyright. Users utilize these models at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The models have been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of these models.