Modelpublic

minpeter/LoRA-FLUX-anime-style-v2

sourceHugging Faceotherupdated 1y agoView on Hugging Face

10likes336downloads

Model Card

Fantasy Anime Style

A Flux LoRA trained on a local computer with ostris/ai-toolkit.

Fixed certain character style biases that existed in FAS v1. It still has some difficulties generating images that are not people/food, but I hope this will be fixed in the next version so that it can be used as a "style LoRA".

guidancescale 3.5, adapterweights 0.7~1.0 is recommended. Feel free to adjust the rest and test it out :)

Trigger words

You should use anime style to trigger the image generation.

Creating images using a diffusers

import torch
from diffusers import AutoPipelineForText2Image

pipe = AutoPipelineForText2Image.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16).to("cuda")

pipe.unload_lora_weights()
pipe.load_lora_weights("minpeter/FAS-v2", weight_name="adapter.safetensors", adapter_name="fas-v2")
pipe.set_adapters(["fas-v2"], adapter_weights=[0.8])

prompt = "anime style, Digital illustration with a glamorous and captivating atmosphere. Attractive figure with stylish short black bob and bangs. Eye-catching composition showing a beautiful side profile with head turned to the right from the back. Wearing large, sparkling, ornate gold earrings and an elegant lace choker that accentuates the neckline. In the background is a huge, brightly shining mysterious orange full moon dominating the night sky. Dramatic yet warm lighting gently envelops the figure and adds a dreamy feel."

image = pipe(
    prompt=prompt,
    guidance_scale=3.5,
    height=1024,
    width=1024,
    num_inference_steps=15,
    # generator=torch.Generator(device="cpu").manual_seed(208070), # for reproducibility
).images[0]

image.save("s4.png")