Modelo de Consistencia Latente (LCM) LoRA: SDXL

LCM-LoRA está soportado en la biblioteca 🤗 Hugging Face Diffusers a partir de la versión v0.23.0 en adelante. Para ejecutar el modelo, primero instala la última versión de la biblioteca Diffusers, así como peft, accelerate y transformers.

pip install --upgrade pip
pip install --upgrade diffusers transformers accelerate peft

Text-to-Image

El adaptador se puede cargar con su modelo base stabilityai/stable-diffusion-xl-base-1.0. Luego, el programador necesita ser cambiado a LCMScheduler y podemos reducir el número de pasos de inferencia a solo 2 a 8 pasos.

import torch
from diffusers import LCMScheduler, AutoPipelineForText2Image

model_id = "stabilityai/stable-diffusion-xl-base-1.0"
adapter_id = "latent-consistency/lcm-lora-sdxl"

pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch.float16, variant="fp16")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
pipe.to("cuda")

# cargar y fusionar lcm lora
pipe.load_lora_weights(adapter_id)
pipe.fuse_lora()

prompt = "Autorretrato como pintura al óleo, un hermoso cyborg con cabello dorado, 8k"

# desactivar guidance_scale pasando 0
image = pipe(prompt=prompt, num_inference_steps=4, guidance_scale=0).images[0]

Inpainting

LCM-LoRA también se puede usar para el 'inpainting'.

import torch
from diffusers import AutoPipelineForInpainting, LCMScheduler
from diffusers.utils import load_image, make_image_grid

pipe = AutoPipelineForInpainting.from_pretrained(
    "diffusers/stable-diffusion-xl-1.0-inpainting-0.1",
    torch_dtype=torch.float16,
    variant="fp16",
).to("cuda")

# configurar el programador
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")
pipe.fuse_lora()

# cargar imagen base y máscara
init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").resize((1024, 1024))
mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").resize((1024, 1024))

prompt = "un castillo en la cima de una montaña, altamente detallado, 8k"
generator = torch.manual_seed(42)
image = pipe(
    prompt=prompt,
    image=init_image,
    mask_image=mask_image,
    generator=generator,
    num_inference_steps=5,
    guidance_scale=4,
).images[0]
make_image_grid([init_image, mask_image, image], rows=1, cols=3)

Combinar con LoRAs estilizados

LCM-LoRA se puede combinar con otros LoRAs para generar imágenes estilizadas en muy pocos pasos (4-8). En el siguiente ejemplo, utilizaremos el LCM-LoRA con el LoRA 'papercut'.

import torch
from diffusers import DiffusionPipeline, LCMScheduler

pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    variant="fp16",
    torch_dtype=torch.float16
).to("cuda")

# configurar el programador
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# cargar LoRAs
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl", adapter_name="lcm")
pipe.load_lora_weights("TheLastBen/Papercut_SDXL", weight_name="papercut.safetensors", adapter_name="papercut")

# combinar LoRAs
pipe.set_adapters(["lcm", "papercut"], adapter_weights=[1.0, 0.8])

prompt = "papercut, un lindo zorro"
generator = torch.manual_seed(0)
image = pipe(prompt, num_inference_steps=4, guidance_scale=1, generator=generator).images[0]
image

ControlNet

import torch
import cv2
import numpy as np
from PIL import Image

from diffusers import StableDiffusionXLControlNetPipeline, ControlNetModel, LCMScheduler
from diffusers.utils import load_image

image = load_image(
    "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_image_vermeer.png").resize((1024, 1024))

image = np.array(image)

low_threshold = 100
high_threshold = 200

image = cv2.Canny(image, low_threshold, high_threshold)
image = image[:, :, None]
image = np.concatenate([image, image, image], axis=2)
canny_image = Image.fromarray(image)

controlnet = ControlNetModel.from_pretrained("diffusers/controlnet-canny-sdxl-1.0-small", torch_dtype=torch.float16, variant="fp16")
pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    controlnet=controlnet,
    torch_dtype=torch.float16,
    safety_checker=None,
    variant="fp16").to("cuda")

# configurar el programador
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")
pipe.fuse_lora()

generator = torch.manual_seed(0)
image = pipe(
    "imagen de La Mona Lisa",
    image=canny_image,
    num_inference_steps=5,
    guidance_scale=1.5,
    controlnet_conditioning_scale=0.5,
    cross_attention_kwargs={"scale": 1},
    generator=generator,
).images[0]
make_image_grid([canny_image, image], rows=1, cols=2)

T2I Adapter

Este ejemplo muestra cómo utilizar el LCM-LoRA con el Canny T2I-Adapter y SDXL.

import torch
import cv2
import numpy as np
from PIL import Image

from diffusers import StableDiffusionXLAdapterPipeline, T2IAdapter, LCMScheduler
from diffusers.utils import load_image, make_image_grid

# Prepare image
# Detect the canny map in low resolution to avoid high-frequency details
image = load_image(
    "https://huggingface.co/Adapter/t2iadapter/resolve/main/figs_SDXLV1.0/org_canny.jpg").resize((384, 384))

image = np.array(image)

low_threshold = 100
high_threshold = 200

image = cv2.Canny(image, low_threshold, high_threshold)
image = image[:, :, None]
image = np.concatenate([image, image, image], axis=2)
canny_image = Image.fromarray(image).resize((1024, 1024))

# cargar el adaptador
adapter = T2IAdapter.from_pretrained("TencentARC/t2i-adapter-canny-sdxl-1.0", torch_dtype=torch.float16, varient="fp16").to("cuda")

pipe = StableDiffusionXLAdapterPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    adapter=adapter,
    torch_dtype=torch.float16,
    variant="fp16", 
).to("cuda")

# configurar el programador
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")

prompt = "Hada mística en el bosque, mágica, imagen 4k, alta calidad"
negative_prompt = "dedo extra, menos dedos, cortado, peor calidad, baja calidad, defecto, deformado, mutado, feo, desfigurado"

generator = torch.manual_seed(0)
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    image=canny_image,
    num_inference_steps=4,
    guidance_scale=1.5, 
    adapter_conditioning_scale=0.8, 
    adapter_conditioning_factor=1,
    generator=generator,
).images[0]
make_image_grid([canny_image, image], rows=1, cols=2)