1 Star 0 Fork 0

db/a1111-stable-diffusion-webui-vram-estimator

加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
克隆/下载
贡献代码
同步代码
取消
提示: 由于 Git 不支持空文件夾,创建文件夹后会生成空的 .keep 文件
Loading...
README
MIT

VRAM Estimator for stable-diffusion-webui

When using stable-diffusion-webui, if you find yourself frequently running out of VRAM or worried that pushing your settings too far will break your webui, this extension might be of use. It gathers a set of statistics based on running txt2img and img2img with various different settings and uses extrapolation to estimate the amount of VRAM your settings will use.

Preview

Usage

First you will need to gather some data from running a bunch of image generations for the extension to use (you'll only have to do this step once). Go to the VRAM Estimator tab and set the Max Image Size and Max Batch Count parameters to the maximum that your system can handle when generating with txt2img and Hires Fix enabled (Hires Fix uses more VRAM than plain txt2img even when the target resolution is the same). The default settings of size 960 and 8 batch count are appropriate for an RTX 3090. Make sure you've closed any other programs that can take up VRAM, then click Run Benchmark and watch the console for progress. The benchmark will generate several images at different resolutions with txt2img and img2img and save the VRAM used by each run.

After the benchmark has finished, you can go to the txt2img or img2img tab and adjust the sliders for Width/Height/Hires Fix to see the estimated VRAM usage. Note that the estimate is based on the amount of used VRAM on your system at the time you change the slider. As a rule of thumb you should expect about 1 GB more of VRAM to actually be used by the generating operation you choose.

Also note that this extension does not take into account VRAM allocated by other extensions like ControlNet; it only looks at the width, height and batch count of the vanilla generation config. If you update webui, it's worth running the benchmark again if you find the estimates are no longer accurate.

MIT License Copyright 2023 space-nuko Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

简介

在逐渐增加的尺寸和批处理大小下运行 txt2img、img2img、highres-fix 直到 OOM,并将数据输出到图表。 展开 收起
取消

发行版

暂无发行版

贡献者

全部

语言

近期动态

不能加载更多了
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化
1
https://gitee.com/dbscholar0/a1111-stable-diffusion-webui-vram-estimator.git
git@gitee.com:dbscholar0/a1111-stable-diffusion-webui-vram-estimator.git
dbscholar0
a1111-stable-diffusion-webui-vram-estimator
a1111-stable-diffusion-webui-vram-estimator
master

搜索帮助