Comments (4)
I don't know what's causing issue but you got to realize 100% is not 100%
the progress while we showing is only the sampling steps
but there's also other things that happen after
such as the VAE or potentially other post processing
so the true progress is 100%+vae+other post processing
so if you "preemptively stops 1 step before the final"
all that dose is make 100%+vae+other post processing
in to 95%+vae+other post processing
if it gets stuck afterwards like in VAE, it's still stuck
I'm guessing the issue is during VAE your system doesn't have enough vram and start using system ram which causes massive slow down
see #11063 (comment)
live preview works differently from normal image, (it essentially decodes the latent space using a different method by default using a cheaper method so it's quicker)
if the issue is with VRAM usage during VAE pahse
there's a couple of things you can try
you can try using TAESD as the final VAE, using TAESD will result in image quality drop but should be much faster and use less VRAM
or you can try https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
this can reduce the vram usage during VAE pahse
if the issue is it getting stuck during other post processing then more investigation needs to be done
you should really fill in the issue report with the proper information we requested
like the Sysinfo
, if you uploaded that then I can see your settings and what system you're on what graphics card you're using
which may give us clue on what is causing the issue
from stable-diffusion-webui.
@w-e-w Updated Sysinfo
Just checked a few generations, vram is around 4.6~/6.0 during hires then spikes to 5.7~/6.0 on final processing for a 576x640 image upscaled by 2x.
from stable-diffusion-webui.
as expected RTX 2060 with 6GB vram
if it spikes up to 6 then then it's likely that it's actually going into System RAM which can cause slow down
so I think it's highly likely it's caused by not enough VRAM
if you want to test i if this is the case you can you fellow #11063 (comment)
and set CUDA - Sysmem Fallback Policy
.to Prefer No Sysmem Fallback
this ment the driver will not try to use system RAM but just ERROR when not enouth VRAM (image gen will fail)
it is possible in your situation --medvram
or --lowvram
may help
enableing FP8 weight
in setting may also help
basically all the settings that reduces vram use may help
from stable-diffusion-webui.
@w-e-w I don't know if it's actually a vram issue. For a normal 50-90 second gen the vram will spike to 5.8~ on final and normal ram will bump up a bit. I just got a generation that took 8+ minutes but it also only spiked to 5.8 and bumped up normal ram a little, so there's no difference there besides the time it takes to finish. Between the 8 minute gen and the gen before it which took 54 seconds, the only difference was the seed and a slight change made to the prompt, the size and everything else was kept the same.
I followed the instructions and turned CUDA - Sysmem Fallback Policy but it didn't do anything different when generating. I even tried making a couple images twice as large as I normally do, capped at vram 5.8 and uses a bunch of ram but no delay and no failure.
from stable-diffusion-webui.
Related Issues (20)
- [Bug]: Torch is not able to use GPU HOT 1
- [Bug]: Not working all HOT 2
- [Bug]: HOT 2
- [Bug]: Loras not working aymore HOT 1
- Update Linux Install Instructions HOT 3
- [Bug]: ONNX Runtime Fails to Load CUDA Provider HOT 2
- [Bug]: Cannot start stable diffusions HOT 3
- [Bug]: New update broke LoRA's on apple silicon HOT 1
- [Bug]: Checkpoint isnt working and the lora are empty HOT 1
- [Bug]: partially initialized module 'modules.sd_samplers_common' has no attribute 'SamplerData' (most likely due to a circular import) HOT 1
- [Bug]: cpu没有全部投入图生图
- Can't start Forge's run.bat HOT 1
- [Bug]: When I set the face model of the reactor function of the comfy ui to GFPGAN, it does not come out normally. HOT 6
- [Bug]: getting an import error about can't import name packaging from pkg_resources HOT 7
- [Fix Guide] ImportError: cannot import name 'packaging' from 'pkg_resources'
- [Feature Request]: Should be easy enough to add a temperature setting with latest pytorch update
- [Bug]: Failed to run on Windows 11 HOT 1
- [Bug]: Error code 1 ( windows 10, nvidia ) HOT 15
- [Bug]: Windows, ctrl + c to quit run.bat leaves cd at ?/installdir/webui instead of ?/installdir
- [Feature Request]: What if I modify the parameters in the configuration file so that the program can render images through the CPU? HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from stable-diffusion-webui.