Cannot start fast stable diffusion notebook
It was fine this morning but now im getting error messages when starting the pod such as invalid username and password. When following the link to the LastBen repo on huggingface it seems to no longer exist.
150 Replies
Pod ID is 9ccqfpgn4lvxlf
Solution
Template has been discontinued/abandoned. RunPod should delist it.
ok cheers
what is the best alternative to fast stable diffusion?
This one has latest A1111, ControlNet, ADetailer etc
Is there any faster way? this template is up for 20 minutes and the A1111 port is still not up yet
It only has to sync first time, doesn't have to sync after that
I assume you are using network storage?
actually I don't use network storage. I start all over again every time
I wonder why its taking so long to sync then.
I had this problem with other templates as well. for ULTIMATE Stable Diffusion Kohya ComfyUI InvokeAI I had to wait about 30 minutes
That one has a lot of stuff that nees to be synced, the A1111 only has A1111 and a few extensions, so should not take the same amount of time to sync.
I try again
Yeah, for the first time sync it will take longer
Next time you load, it will be faster because some of it are installed already
Its very quick to sync in community cloud, I don't know why secure cloud syncs so slowly even without network storage, thats a mystery to me.
Might be from the filesystem in secure xloud
i use secure cloud...maybe thats the problem
Why??? It was my favourite template!!! How can I train Loras now? I don´t understand why we don´t have any notification weeks before.
You can use Kohya_ss for training Lora.
Kohya_ss is much better nowdays
for DB would go with #Better EveryDream2trainer - CUDA12
Kohya is sooo slow... 😦
It depends on your settings
Which top 3 affects the most
Tried EveryDream?
Pretty annoyed that my favourite template was removed. A1111 Stable Diffusion 1.9.4 is an okay replacement, but does it have any way to see the Python outputs, like you could with Fast Stable Diffusion? There's information in there that I'm just not getting now.
I think you can use jupyter or web cli
To access the logs somewhere inside /workspace
It should be .txt or .log
I think there is a folder called output within /workspace-stable-diffusion-webui where you can find the output files
It does all into is saved to log file
What exacly things you liked about fast SD?
how can i restore Huggingface A1111 backup file with this template?
or have any backup/restore method for A1111?
Yep there /
Yeah, I've found the logs, it'll do.
You probably need restore backup manually
I'm sure it all depends on what you're used to. But I had the scripts customized to my liking and it was nice to have everything I needed (except the webui itself) in that one Jupyter window. The community template just feels cumbersome by comparison.
What exacly customized?
GitHub
GitHub - ashleykleynhans/a1111-huggingface-sync: Synchronises A1111...
Synchronises A1111 models with a Huggingface repository - ashleykleynhans/a1111-huggingface-sync
Just little things like turning on xformers (for which I now have to manually go into the files each time I start a new pod) and automatically downloading all my go-to models. (The download script still works, but now I have to go out of my way to run that, while it was previously part of the start-up script.) None of it really matters, it’s just mildly annoying.
My template has xformers preinstalled and you can store models on NS
I could've sworn I had issues until I manually enabled xformers, but maybe I broke something when starting it up.
Why use xformers
It's depends on template
The A1111 template above also has xformers installed already so you don't need to install it unless you installed an extension that broke everything
A1111 extensions can sometimes be like Crowdstrike, but breaking A1111 installations instead of Windows machines 😆
Ah this is terrible https://github.com/TheLastBen/fast-stable-diffusion/issues/2887 been using it daily
GitHub
RunPod not showing Jupyter Lab option · Issue #2887 · TheLastBen/fa...
The runpod template isn't providing the option connect to jupyter lab anymore.
Yeah, some community members create templates out of the goodness of their hearts, but unfortunately others only do it for profit if the hosting provider pays them for it.
Yeah i mean its a bummer, we will need to move a few companies now to paperspace i guess
if you want you can explore / try other trainers( more listed on above chats) or webui like a1111 or fooocus
im sure they're amazing too
Depends on whether its for training or inference, Fast SD template did both
It seems more people were using it for training than for inference
ic, yeah
i would need something for comfyUI, any good for that?
What do you mean
What do you need
template that has a comfyUI inside
actually found one - ComfyUI - AI-Dock
There are a few different ComfyUI templates. Papa Madiator has a "Better ComfyUI" template as well.
Depends on your personal preferences I guess.
I would prefer a one that is not gonna go away 😄
Yeah use madiator's then
Papa Madiator is employed by RunPod so his are probably the least likely to disappear
Hahahaha
ok 😄
I also don't think the other community templates are "under contract" so the developers don't really need to be spiteful and remove them
Yep
I think it was a bad idea to make it that way, because I mean templates are publicly available code and should be kinda opensource
I think that was one of the first, maybe even the first community template, so RunPod probably had to incentivize people to create community templates at the time, but now there are literally dozens to choose from so they don't need to do it anymore.
Yeaa
Anyways it's done
Yes, but if RunPod is earning and getting new customers because of the SD-Fast template and if the template has UVP, in that case, it makes sense - I understand the "contract" written templates, but the situation right now is shitty for many customers that pay runpod.
I used the template because it was published on the Blog
Somehow maybe it's not really a great deal after some time maybe there were other factors too beside that
Yeah ic many doesn't like this decision too because they lose the usual fast sd
hey! where can I find those trainers under A111 and Forge? Thanks!
would not recomend training with auto111 or forge
Check this message
This is what I mean in chats above
Yeah those are meant for generation only, unless you got some custom extensions
Try those which is talked about in above chats
A1111 has a dreambooth extension but nobody really uses it anymore, they use things like Kohya, EveryDream2Trainer, OneTrainer etc instead.
Ooh
I only used DB extension once deleted it very fast
Loll
Why
It sucks?
Yes
Too many options to choose from
any suggestion on how to move from fast sd to other templates like a1111 or forge? I tried to deploy a pod with a1111 template but it's syncing venv like forever
why there is no need to do venv syncing for fast sd?
Just the first time
Maybe everything is in the template or it has better design pattern
totally agreed. SD-fast was very easy to use and we have been using it for a long time. Now we are fucked and need to fix this asap
It pulls files from Hugging face etc
yeah maybe thats why its faster
ok finally it's working.... all you need is patience
I got @Papa Madiator Better ComfyUI working yesterday, and it works well. I am happy that I don't need to see the Jupyther anymore it was terrible, but I did lost a day and a lot of nerves to move my out setup to new setup.
@jakethesnake1708 better idea just make something better that foes not depend on some external packages
Thanks for feedback
you mean that you are training in kohya_ss, right?
what is working? fast SD template again?
Please, say yes!!
wait its working?
not sure...
We have to restore fast stable diffusion template, boys 😄
I miss it too
no it's not working, and as long as I understand the conversation in the GitHub it won't
to clarify it takes time to load models, setup custom nodes, etc.
no not the fast SD template. We have moved to another one
ok! wich one? can you share with us? thanks!!
I am using this one ashleykza/a1111:1.9.4.post7
was thinking about moving to comfyUI too, but right now we rely heavily on the sd webui api, does comfyUI api functions very similar to a1111 webui api?
No, its very different i'd say
ComfyUI has an API and you submit your workflow to the API
The web ui can generate the JSON for you to subit to the API which makes ComfyUI actually better than A1111
And ComfyUI is more powerful, more flexible and gets new features long before A1111
ComfyUI is maintained by Stability AI, where A1111 is maintained solely by the community
Hmm really?
Yes
i think stability ai isn't gonna maintain ComfyUI anymore
But, im sure it'll still be well maintained
Why not? comfyanonymous developer is employed by them
Yeah, was*
What do you mean was?
Well not anymore now
Where did you get this info?
Wherever i go in my internet session, search it up or join communities cmon.
What internet session? Please stop being sarcastic, I am already a member of communities and never saw this information, I don't understand how someone with your attitude becomes a "community helper" when you just want to be sarcastic and not help anyone at all.
Maybe @Madiator2011 will know since he is actually interested in helping people.
hmm let me see
Comfy Org
Comfy Org
Creators of ComfyUI. We are a team dedicated to iterate and improve ComfyUI, support the ComfyUI ecosystem with tools like node manager, node registry, cli, automated testing, and public documentation.
I kinda do not like auto1111 cause poor extension managment
pls ah, i feel very blamed if you say im sarcastic, im just trying to give informations i've heard before
but if you want to find sources try to find it yourself and stop forcing me to
for me comfy ui works much much better
yeah me too
more tools, but has some learning curves too if you're moving from a1111
I simply asked where you got the information, so you could have just answered it without saying "cmon" etc , that was compltely uncalled for and unnecesary
okay no "cmon", read this up https://blog.comfyui.ca/comfyui/update/2024/06/18/Next-Chapter.html
ComfyUI blog
The next chapter for ComfyUI
As some of you already know, I have resigned from Stability AI and am starting a new chapter. I am partnering with mcmonkey4eva, Dr.Lt.Data, pythongossssss, robinken, and yoland68 to start Comfy Org. We will continue to develop and improve ComfyUI with a lot more resources.
Thank you, thats all I wanted to know
Wow, interesting
can you share the runpod template link? Thanks!!
https://www.runpod.io/console/explore/ts8ze6urzh
this one, right?
but you can´t train loras...
This one can train Lora, Dreambooth, full fine tuning etc
thanks!
My problem is... kohya is too complicated for trainings... Fast SD was so simple. Do you know any other way like that?
Kohya is pretty simple once you get to understand it
I wasnt fan of Kohya too but now I use it daily
There are even some YouTube videos showing how to use it
really? I can´t get the same results than fastsd
But i will try it!
thanks so muchç
Very annoyed that Fast SD was removed without any prior notice. it was the only template that worked for me, now I dont have any A1111 options as each is configured differently and takes ages to start. Any possibility of bring back thr Fast SD?
It can't be brought back, TheLastBen pulled the code from his repos, so its his decision not RunPod's decision
yep the maintainer closed the support for runpod
Well I cannot get any other A1111 instances to work - errors on all of them.
Then you're likely doing something wrong, otherwise other people would have errors too.
I have submitted the error to ashleyk on git
cant agree more. Removing the template without any prior notice is pretty bad, affecting so many apps/services that were built on Runpod using the fast SD template. RP should have done a better job communicating this
In short: Runpod didn't remove the fast SD template, the author of the "closed source" template was actually paid rent by runpod and he decided to make the template stop working for Runpod users because he stopped getting paid..cmiiw.
so runpod should let us know if they stopped paying fast SD template and there might be risks of fast sd template not supported in the near term
Yes correct, that would be better but maybe it went unannounced
also its a community template isnt it
wasn't it official in the beginning?
Hmm not sure, i mean its not maintained by runpod
anyways what is done is done already, there is some alternatives you can use for now
sure man, I am not complaining much. more like advice to runpod operations. I have been liking runpod service for quite some time and I am hoping that it continues to develop well
ic yeah sure
me too
bc if otherwise we might have to find a totally different infrastructure to work on, which is a lot a lot more hassles
It was maintained by community but listed as an "official" template so yeah there should have been some communication about it, I agree.
I have not been able to find another A1111 template that works. Ashleyk said that the error I received in the new A1111 template is a bug for that version. Does anyone know of a template that works on RunPod? I have tried a number of templates but they are not maintained and after an hour of waiting for the container logs I exit. More $ wasted on trying to run templates that should not be available if they do not work or are not maintained.
I'v just pushed #Better ComfyUI - CUDA12 with video tutorial
If there is a bug in A1111, log it in the A1111 Github repo:
https://github.com/AUTOMATIC1111/stable-diffusion-webui
You can also set the docker image to an earlier version if its an issue in the latest version of A1111.
GitHub
GitHub - AUTOMATIC1111/stable-diffusion-webui: Stable Diffusion web UI
Stable Diffusion web UI. Contribute to AUTOMATIC1111/stable-diffusion-webui development by creating an account on GitHub.
man it seems like AshleyK is not here anymore, was going to ask a question about his template...Do i need to keep the venvs folder once the sync is done?
If you delete it, it will have to sync again, why would you want to remove it?
but my sd webui is already installed and running. If this folder is no longer needed I would like to free up some space
they part of docker image so even if you delete them from container storage they will be back
I am using network storage
and?
You can't delete venv
application won't work anymore if you delete it
All the dependencies for the app are installed in the venv
Well I mean you can delete it, but you shouldn't delete it otherwise everything will break
well this folder takes up 17G!
Yeah dependencies are large especially torch
disk space is cheap compared to GPU, add space if you need more, don't try and delete venv
It's like saying I want to install new game let me delete windows so I can save space for my game 😄
gocha thanks. Didn't have this issue before when using the fast SD template so I am confused
it was most likelly downloading venv from external page
Why is Better "ComfyUI CUDA12 - Outdated" now outdated? 😏 Just installed it last week..
Maybe there is a anew version for it?
There are constantly new updates being pushed to ComfyUI GitHub, sometimes even several times a day
@Perspectalog it's cause main templates are Light and Full versions
squashed some of bugs
Thanks!