22/10/08更新,部署了一下原版网页,点击登录即可玩,显卡数量有限人多了可能会爆炸或者很慢,耐心等等,不行就f5
22/10/08更新:替换失效的奶牛快传链接到onedrive
硬件需求:
一台拥有一张至少有11G 显存的NVIDIA GPU的linux系统的x86设备。
软件需求:
NVIDIA驱动(CUDA 11.6 Toolkit)
Docker 19+
nvidia-container-toolkit
准备工作:
- 安装docker
该命令国内访问较慢,可以查国内镜像安装
curl -fsSL https://get.docker.com | bash
2. 安装nvidia-container-toolkit
Ubuntu, Debian:
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add –
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker
CentOS/RHEL
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.repo | sudo tee /etc/yum.repos.d/nvidia-docker.repo
sudo yum install -y nvidia-container-toolkit
sudo systemctl restart docker
3. 安装显卡驱动
这个略过
4. 确认显卡驱动已经安装好, nvidia-smi可以看到显卡
5. 确认nvidia-container-toolkit安装成功
docker run –help | grep -i gpus
6. 下载NovelAI模型相关文件
7. 解压NovelAI模型相关文件
tar -zxvf novelai.tar.gz
8. 下载docker镜像相关文件
9. 导入Docker镜像
docker load -i novelaidocker.tar
10. 运行Docker
(如果希望包含nsfw内容,则把-e MODEL_PATH=”/root/stableckpt/animesfw-latest” 改成 -e MODEL_PATH=”/root/stableckpt/animefull-latest”)
解压的novelai位置替换为你实际解压NovelAI模型相关文件出来的novelai文件夹的位置,如/root/novelai
docker run –gpus all -d -p 80:80 -v 解压的NovelAI位置:/root -e DTYPE=”float32″ -e AMP=”1″ -e MODEL=”stable-diffusion” -e DEV=”True” -e MODEL_PATH=”/root/stableckpt/animesfw-latest” -e MODULE_PATH=”/root/stableckpt/modules” -e TRANSFORMERS_CACHE=”/root/transformer_cache” -e SENTRY_URL=”” -e ENABLE_EMA=”1″ -e VAE_PATH=”/root/stableckpt/animevae.pt” -e BASEDFORMER=”1″ -e PENULTIMATE=”1″ novelai:latest gunicorn main:app –workers 1 –worker-class uvicorn.workers.UvicornWorker –bind 0.0.0.0:80
11. 查看容器状态
查询出容器ID
docker ps
docker logs [Container ID]
出现“Application startup complete.”即代表程序已经就绪
docker attach [Container ID]
attach进入docker可以看到当前任务实时的生成进度
11. 请求API
参考代码,具体参照leak的前端后端项目,以及其中的sd-private\hydra-node-http\main.py
prompt中masterpiece, best quality, 开头对应原版web Add Quality Tags选项,不建议删除,后面直接跟自己prompt即可
uc部分对应web Undesired Content,建议保留默认
sampler是采样方法,可选plms/ddim/k_euler/k_euler_ancestral/k_heun/k_dpm_2/k_dpm_2_ancestral/k_lms
seed是种子,自己随机一个整数数字,不然一直会出一样的结果。
n_samples代表要生成几张图片
import requests
import json
import base64
import random
endpoint = “http://10.10.12.67/generate”
data = {“prompt”: “masterpiece, best quality, brown red hair,blue eyes,twin tails,holding cat”, “seed”: random.randint(0, 2**32)
,”n_samples”:1,”sampler”:”ddim”,”width”:512,”height”:768,”scale”:11,”steps”:28,”uc”:”lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry”}
req = requests.post(endpoint, json=data).json()
output = req[“output”]
for x in output:
img = base64.b64decode(x)
with open(“output-” + str(output.index(x)) + “.png”, “wb”) as f:
f.write(img)
生成的效果图示例
请登录后查看评论内容