anzu_flux v2.2のLoRAが公開されていたので使ってみる

huggingface.co

Pythonスクリプト

import torch 
from diffusers import FluxPipeline
import gc

def flush():
    gc.collect()
    torch.cuda.empty_cache()

model_id = "black-forest-labs/FLUX.1-dev"

prompt="a photo of f5h8_woman holding a sign that says 'I LOVE LoRA!"

pipeline = FluxPipeline.from_pretrained(
        model_id,
        transformer=None,
        vae=None
).to("cuda")

with torch.no_grad():
    prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(
        prompt=prompt,
        prompt_2=None,
        #max_sequence_length=256
    )

del pipeline
flush()

pipeline = FluxPipeline.from_pretrained(
    model_id,
    text_encoder=None,
    text_encoder_2=None,
    tokenizer=None,
    tokenizer_2=None,
    torch_dtype=torch.bfloat16
)

pipeline.load_lora_weights("anzu-flux-LoRA_v22.safetensors")

pipeline.enable_sequential_cpu_offload()

seed = 12345
generator = torch.Generator().manual_seed(seed)
image = pipeline(
    prompt_embeds=prompt_embeds.bfloat16(),
    pooled_prompt_embeds=pooled_prompt_embeds.bfloat16(),
    width=1024,
    height=1024,
    num_inference_steps=27,
    generator=generator,
    guidance_scale=3.5,
    joint_attention_kwargs={"scale": 1.0},
    
).images[0]

image.save(f"lora_result_seed{seed}.jpg")

結果