Stable Diffusion（Huggingface的方式安装）

Huggingface的方式安装

在Huggingface个人目录下有一个token号，这个tokens号要在服务器登陆的过程中进行添加；

1，在服务器登陆要输入huggingface登陆：

huggingface-cli login
# READ 权限的就够了

运行完命令，按照提示将之前的token复制过来，然后 Y ，回车，登陆成功！

2，相关包安装

pip install --upgrade diffusers transformers scipy
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
pip install accelerate # 这个加速包可以不装，但是他一直提示看着难受的就装了

3，运行模式

import torch
from torch import autocast
from diffusers import StableDiffusionPipeline

model_id = "CompVis/stable-diffusion-v1-1"
device = "cuda"


pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
pipe = pipe.to(device)

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["sample"][0]  
    
image.save("astronaut_rides_horse.png")

哈哈，报错，error key ‘sample’
打印下pipe看看

只有两个key：images 和 nsfw_content_detected
那改成 images试试

import torch
from torch import autocast
from diffusers import StableDiffusionPipeline

model_id = "CompVis/stable-diffusion-v1-1"
device = "cuda"


pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
pipe = pipe.to(device)

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["images"][0]  
    
image.save("astronaut_rides_horse.png")

第一次运行会下载模型，需要点时间，生成的时候显存大概占用12G

成功了，看下生成的怎么样。

马马虎虎，还有很大优化空间。

补充

显卡显存不足10G的话可以改成float16 precision

import torch
from torch import autocast
from diffusers import StableDiffusionPipeline

model_id = "CompVis/stable-diffusion-v1-1"
device = "cuda"

pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
#pipe = StableDiffusionPipeline.from_pretrained(model_id, use_auth_token=True)
pipe = pipe.to(device)

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["images"][0]  
    
image.save("astronaut_rides_horse.png")

#To swap out the noise scheduler, pass it to from_pretrained:
要交换噪声调度器，将其传递给from_pretrained：

import torch
from diffusers import StableDiffusionPipeline, LMSDiscreteScheduler

model_id = "CompVis/stable-diffusion-v1-1"
# Use the K-LMS scheduler here instead
scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, use_auth_token=True)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["images"][0]  
    
image.save("astronaut_rides_horse.png")

哈哈啊哈啊
PS：huggingface-cli: command not found

参考

CompVis/stable-diffusion-v1-1（Hugging Face）
模型方法–Stable Diffusion
CompVis/stable-diffusion（github）
Stable Diffusion with 🧨 Diffusers