FLUX.1-dev-LoRA-Ghibli

FLUX.1-dev-LoRA-Ghibli

This is a Ghibli style LoRA trained on FLUX.1-dev.

<Gallery />

Trigger words

You should use ghibli style to trigger the image generation.

Inference

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.load_lora_weights("InstantX/FLUX.1-dev-LoRA-Ghibli", weight_name="ghibli_style.safetensors")
pipe.fuse_lora(lora_scale=1.0)
pipe.to("cuda")

prompt = "ghibli style, a shepherd boy floating on a wooly cloud-whale, holding a glowing dandelion staff to guide sheep-shaped cumulus, miniature storm clouds grazing nearby, his patched jacket flapping in high-altitude winds, aurora-like ribbons in peach and lavender stretching across the sky"

image = pipe(prompt, 
             num_inference_steps=24, 
             guidance_scale=3.5,
             width=960, height=1280,
            ).images[0]
image.save(f"example.png")