margin2model/Krea-2-lora-test

Krea 2 LoRA

This Diffusers LoRA was trained on krea/Krea-2-Raw using the facebook/winoground dataset. Krea recommends training adapters on RAW and loading them on the fast distilled krea/Krea-2-Turbo pipeline for eight-step inference.

import torch
from diffusers import Krea2Pipeline

from huggingface_hub import hf_hub_download

lora_path = hf_hub_download(
    repo_id="margin2model/Krea-2-lora-test",
    filename="pytorch_lora_weights.safetensors",
)

pipe = Krea2Pipeline.from_pretrained(
    "krea/Krea-2-Turbo",
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()
pipe.load_lora_weights(lora_path)

image = pipe(
    prompt='A cinematic portrait, detailed, soft studio lighting',
    num_inference_steps=8,
    guidance_scale=0.0,
    height=1024,
    width=1024,
    generator=torch.Generator(device="cuda").manual_seed(0),
).images[0]
image.save("image.png")

The adapter is stored as pytorch_lora_weights.safetensors in Diffusers-native Krea 2 format.

License

This adapter is a derivative of Krea 2 and is subject to the Krea 2 Community License and the Krea Acceptable Use Policy. The community license permits commercial use only below its stated annual revenue threshold; other commercial use requires an enterprise license from Krea. Deployers must also implement the safeguards required by that license.