Krea 2 Realism LoRA

A realism LoRA for Krea 2. It pushes Krea 2 toward natural, candid, true-to-life photography — everyday people, places, and light — without needing any "photorealistic / 8k / DSLR" style keywords. Just write a normal, descriptive sentence and let the LoRA do the rest.

Trained with the FAL Krea LoRA Trainer on krea/Krea-2-Raw and meant to be used directly on krea/Krea-2-Turbo (per Krea's recommended workflow of training on Raw and running on Turbo).

Trigger words

No trigger word is required. Simply describe the scene in plain natural language. Longer, descriptive single-sentence prompts work best, e.g.:

A young woman taking a selfie in front of a steamy bathroom mirror with her phone partly covering her face, soft morning light coming through a frosted window, water droplets on the glass and a white towel wrapped around her hair

Usage with 🤗 diffusers

Install diffusers from source (for Krea2Pipeline):

pip install git+https://github.com/huggingface/diffusers.git

import torch
from diffusers import Krea2Pipeline

pipe = Krea2Pipeline.from_pretrained(
    "krea/Krea-2-Turbo", torch_dtype=torch.bfloat16
).to("cuda")

pipe.load_lora_weights("gokaygokay/Krea-2-Realism-LoRA")

prompt = (
    "A woman with curly hair laughing while holding an open paperback in a cluttered "
    "secondhand bookshop, tall crowded shelves towering around her in soft warm lamplight"
)

image = pipe(prompt, num_inference_steps=8, guidance_scale=0.0).images[0]
image.save("krea2_realism.png")

Usage with fal

The LoRA can be used directly on the hosted Krea 2 Turbo endpoint by pointing to the weights:

import fal_client

result = fal_client.subscribe(
    "fal-ai/krea-2/turbo/lora",
    arguments={
        "prompt": "A barista with tattooed forearms leaning on the worn wooden counter of a tiny corner cafe holding a cup with delicate latte art, late afternoon sun streaking across shelves of mismatched mugs behind her",
        "loras": [{
            "path": "https://huggingface.co/gokaygokay/Krea-2-Realism-LoRA/resolve/main/krea2_realism_lora.safetensors",
            "scale": 1.0
        }],
        "image_size": {"width": 1280, "height": 1600},
        "num_images": 1
    },
)
print(result["images"][0]["url"])

Recommended settings

• Steps: 8 (Turbo), guidance scale: 0.0
• LoRA scale: ~1.0 (lower to ~0.7-0.8 for a subtler effect)
• Resolution: works well at 1024-1536 on the short side; the examples below are 1280×1600

Examples

All images below were generated on Krea 2 Turbo with this LoRA at scale 1.0, using only plain descriptive prompts (no style keywords).

Download

Weights are available in Files & versions (krea2_realism_lora.safetensors).

Citation and attribution

This model release is maintained by Gökay Aydoğan. If you reference this repository in academic work, please cite it as follows and also cite the upstream models, datasets, or projects it builds upon.

@software{aydogan2026krea_2_realism_lora,
  author = {Aydoğan, Gökay},
  title = {{Krea-2-Realism-LoRA}},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/gokaygokay/Krea-2-Realism-LoRA},
  note = {Model repository; cite the base model and upstream datasets as required.}
}