🌀 Naruto Style LoRA for FLUX.1-dev

This repository contains a LoRA (Low-Rank Adaptation) trained on the FLUX.1-dev base model to generate anime-style images inspired by the Naruto universe.

About

This LoRA brings out vivid Naruto-style line art, character expressions, and backgrounds, making it ideal for stylized image generation tasks. It works seamlessly with the base FLUX model and produces consistent anime-inspired visuals.

Model Description

This LoRA was trained on a tiny subset of lambdalabs/naruto-blip-captions dataset. Only 50 images were used.

Inference Examples

Below is a side-by-side comparison of generations from:

FLUX.1-dev (base model)
FLUX + Naruto LoRA (this repo)

All prompts are in "Naruto style".

Prompt	FLUX.1-dev	FLUX + Naruto LoRA
an old teacher in naruto style
a blonde woman with blue eyes in naruto style
a man in the woods wearing a green jacket in naruto style
a samurai warrior, in naruto style

How to Use

You can use this LoRA with the FLUX pipeline as follows:

from diffusers import FluxPipeline
import torch

ckpt_id = "black-forest-labs/FLUX.1-dev"
pipeline = FluxPipeline.from_pretrained(
    ckpt_id, torch_dtype=torch.float16
)
pipeline.load_lora_weights("zamal/naruto-lora-flux", weight_name="pytorch_lora_weights.safetensors")
pipeline.enable_model_cpu_offload()

image = pipeline(
    "a samurai warrior in naruto style",
    num_inference_steps=28,
    guidance_scale=3.5,
    height=768,
    width=512,
    generator=torch.manual_seed(0)
).images[0]

image.save("samurai_naruto_style.png")

LoRA Training

This LoRA was fine-tuned using the diffusers library with PEFT integration. The dataset included anime-style character references in line with Naruto's visual themes.

zamal
/

naruto-lora-flux