Latent Consistency Model (LCM) LoRA: SDv1-5

LCM-LoRA es compatible con la biblioteca 🤗 Hugging Face Diffusers desde la versión v0.23.0. Para ejecutar el modelo, primero instale 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 SDv1-5 o derivados. En este ejemplo usamos Lykon/dreamshaper-7. Luego, el scheduler debe cambiarse a LCMScheduler y podemos reducir el número de pasos de inferencia a solo 2-8 pasos.

import torch
from diffusers import LCMScheduler, AutoPipelineForText2Image

model_id = "Lykon/dreamshaper-7"
adapter_id = "latent-consistency/lcm-lora-sdv1-5"

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 = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

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

Image-to-Image

LCM-LoRA también se puede aplicar a tareas de imagen a imagen. En este ejemplo utilizaremos el modelo dreamshaper-7 y el LCM-LoRA para stable-diffusion-v1-5.

import torch
from diffusers import AutoPipelineForImage2Image, LCMScheduler
from diffusers.utils import make_image_grid, load_image

pipe = AutoPipelineForImage2Image.from_pretrained(
    "Lykon/dreamshaper-7",
    torch_dtype=torch.float16,
    variant="fp16",
).to("cuda")

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

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdv1-5")
pipe.fuse_lora()

# preparar imagen
url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
init_image = load_image(url)
prompt = "Astronauts in a jungle, cold color palette, muted colors, detailed, 8k"

generator = torch.manual_seed(0)
image = pipe(
    prompt,
    image=init_image,
    num_inference_steps=4,
    guidance_scale=1,
    strength=0.6,
    generator=generator
).images[0]
make_image_grid([init_image, image], rows=1, cols=2)

Inpainting

LCM-LoRA también puede ser utilizado para tareas de inpainting.

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

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

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

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdv1-5")
pipe.fuse_lora()

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

generator = torch.manual_seed(0)
image = pipe(
    prompt="concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k",
    image=init_image,
    mask_image=mask_image,
    generator=generator,
    num_inference_steps=4,
    guidance_scale=4,
).images[0]
make_image_grid([init_image, mask_image, image], rows=1, cols=3)

ControlNet

En este ejemplo, utilizaremos el modelo SD-v1-5 y el LCM-LoRA para SD-v1-5 con canny ControlNet.

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

from diffusers import StableDiffusionControlNetPipeline, 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((512, 512))

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("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    controlnet=controlnet,
    torch_dtype=torch.float16,
    safety_checker=None,
    variant="fp16").to("cuda")

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

# cargar LCM-LoRA
pipe.load_lora_weights("latent-consistency/lcm-lora-sdv1-5")

generator = torch.manual_seed(0)
image = pipe(
    "the mona lisa",
    image=canny_image,
    num_inference_steps=4,
    guidance_scale=1.5,
    controlnet_conditioning_scale=0.8,
    cross_attention_kwargs={"scale": 1},
    generator=generator,
).images[0]
make_image_grid([canny_image, image], rows=1, cols=2)