FINALLY! Stable Diffusion in Colab Notebook for FREE with no disconnects
Вставка
- Опубліковано 15 жов 2024
- Hello, everyone!
What’s happened to Stable Diffusion?
Where to find a colab notebook without disconnects?
How to use Stable Diffusion for free?
How to bypass the Stable Diffusion WebUI block?
What to use instead of Google Colab and Kaggle?
All these questions have become exceptionally relevant in the last weeks.
As Google Colab has blocked almost all possible ways to run Stable Diffusion (Automatic1111, ComfyUI, SwarmUI, etc.), many have been asking, 'Where can we find a Colab notebook without disconnects for free?'
Now, I can provide an answer!
In this video, I will demonstrate the most up-to-date method for running Stable Diffusion with LoRAs, Checkpoints, Embeddings, ControlNet, VAEs for free, and without any limitations using Google Colab's InvokeAI.
Let's get started!
____________________________
Timestamps:
____________________________
🔥 Patreon with Ultimate Drive Colab notebook:
/ marat_ai
🔗 Base Notebook:
/ public-invokeai-90834636
🚨🚨🚨
upd1: If you have an issue with image displaying, the issue has been resolved, and the notebook has just been updated
upd2: If you have an issue with upscalers , the issue has been resolved, and the notebook has just been updated
upd3: If have problem with 'No model', the issue has been resolved, and the notebook has just been updated
Bro, how do you install Vae on this colab? Is there any way?
It doesn't work for me
I have an issue with line 29 in the last cell with the error
"CalledProcessError: Command '['ssh-keygen', '-t', 'rsa', '-b', '4096', '-N', '', '-q', '-f', '/content/id_rsa']' returned non-zero exit status 1."
@@Eltaurus Is not longer availble, Google blocked all stable diffusion collab space, so don't waste your time.
how can i use sdxl on this notebook. it keep disconnecting after few second. only for sdxl, other model are work fine
Good job. Thank you for the Colab Notebook and video. The issue 'No model' message... That's there... It shows me the message and I can't use it. What did I do wrong?
It doesn't work for me, I did everything well, and in the steep 3, it never show me the link to invoke the IA.
Same for me @marat_ai
Amazing video! Just one question, is there a way to use google colab with this code without having to reinstall it each time you want to use it?
Thanks Marat! Just an awkward issue I have here. There are no models available on Invoke. I did download them as explained but nothing. On it for three days now without any result.
You're welcome! I read similar comments today and tested the notebook once again. As I mentioned before, I fixed that issue by adding a cell that downloads the custom model. Please try running all cells, and everything should work properly.
I have used this tool a few times and it worked great without any issues. However, recently I tried to load an SDXL model and it crashed. The system RAM maxed out and Colab disconnected, ending with ^c.
is there any fix.
I wanted to try SDXL checkpoints as well with Invoke AI, funny I didnt had any problems with Juggernaut XL on SD Fooocus.
Hi, in this free model I can send several prompts and create many Excel models?
IT'S STILL WORKING, THAT'S AMAZING
Thank you to share this interesting information. I would like ask how I can install an extension ReActor for Stable Diffusion in your Colab Notebook. This extension is a Fast and Simple FaceSwap
Unfortunately, there is no support for custom extensions yet. Check the workflows; maybe you'll find something useful
Wow, thank you so much! I’ve watched it three times in a row for the promotion 🔥 Love your channel
haha, thanks for your support 🩶
@@marat_aido you have the link?
really cool. i was looking for this for months. Support brother. we need you ❤❤❤❤
did it work for you?
yes. working just fine foe me @@Lauer030
@@Lauer030 doesn't work anymore.
Does it not support extension?
it shows me this error after running the last part: AttributeError: module 'jax.random' has no attribute 'KeyArray'
there's also a problem or bug with step 2, literally none of the models were downloaded like vae, controlnet, etc.
Another great video, signed up to your patreon , do you have to run everything as per your video in the base model everytime I want to login?
Can I use embeddings / negative embeddings in this? How? Please do reply!
Bro can we use Google colab to use for automatic 1111 installing please sir reply 😢
Is there a way to add inpaint or other tools? Would you explain me how?
hello!
it works perfect, but I have an issue - after every generation I have to refresh site to see the results
do you think you can fix that?
This happens to me too since yesterday D:
Wow. Thank you so much. I thought it wasn't working, and I just tried refreshing, and the image was there.
This seems like a problem on Invoke's end, but I'm unsure.
yes, you right, we need wait a few days and then I'll update the notebook
@@marat_ai Hello Marat. Thanks for letting us know.
@@abby_babber Good news!The issue has been resolved, and the notebook has just been updated🫡
Why does it say no models when RV5 is set as true?
Hii, since today the link aint generating on step 3, it shows up as error. Can u fix it please 🙏 😭😭
The final link isn't opening for me. What might be the issue?
awesome. Is there any ControlNet-like stuff in Invoke?
Sure, mentioned in the video
I can't use controlnet, do you know how to fix it? When I import the photo, it says "failed to queue graph"
Bro thankyou for uploading this ❤ i have questions few model doesn't work ex - like juggernaut when i run it invoke gets disconnected how i can fix it.
Probably a dumb question but I'll ask anyway: how do I go about adding a Lora to this? I see the space for it when I open the page, but I just can't seem to do it.
I'm gutted, doesn't work anymore. Everything works perfectly up until I can launch Invoke AI, once in there it's written "No models available" and impossible to generate content. Didn't fiddle with the models file, I just clicked it and leaved it on default. Worked like a charm a week ago.
This time it really doesn't work. Okay, I'll fix everything
having the same issue
@@marat_ai Thanks a lot man, your work is amazing.
You're welcome! The notebook updated!
@@marat_aigetting getting a python error now when loading the code blocks by order
---------------------------------------------------------------------------
CalledProcessError Traceback (most recent call last)
in ()
27 ssh_name = "id_rsa"
28 ssh_path = Path(os.path.dirname(os.getcwd())) / ssh_name
---> 29 gen_key(ssh_path)
30 clear_output()
31
1 frames
/usr/lib/python3.10/subprocess.py in run(input, capture_output, timeout, check, *popenargs, **kwargs)
524 retcode = process.poll()
525 if check and retcode:
--> 526 raise CalledProcessError(retcode, process.args,
527 output=stdout, stderr=stderr)
528 return CompletedProcess(process.args, retcode, stdout, stderr)
CalledProcessError: Command '['ssh-keygen', '-t', 'rsa', '-b', '4096', '-N', '', '-q', '-f', '/content/id_rsa']' returned non-zero exit status 1.
Can you make a similar notebook but for RVC please? (Actually all I need it for is training, so it's even okay if it's without a UI. Most other notebooks (even w/o the UI) appear to be broken - they don't save the final model, - so if that could get resolved by someone I'd be really grateful!
I was thinking of getting into voice cloning, making a couple of tutorials on the topic. In that case, I will definitely create notebooks as well
@@marat_aithat would be a dream come true if rvc runs on sage
Doesn't work for me..
"No models available"
Guys, I'm new to this field. Please tell me why upscale doesn't work? The error "Failed to queue graph", what to do with it?
I've just resolved this issue. Upscalers work once again. Please check it out.
Have you found an alternative to RVC on colab bro?
Is doesnt generate anything for me :/. after launching the invoke ai and setting the models and prompts it is just stuck preparing the image and not doing it.
The issue with the displayed generation results is a consequence of a library update, and it's not related to the notebook. In the next few days, the developers of InvokeAI will fix everything, and I will update the notebook immediately. At the moment, to see the generation results, simply refresh the web page.
Good news!The issue has been resolved, and the notebook has just been updated.
loras I added aren't showing how can I fix it
Thanks for this! I have a question tho. How do I use controlnet openpose in this setup? And how do I also use the png info where I uploas my AI generated image and it will display the prompts used?
You're welcome! To add a model, try editing the YAML file, as most of the models are already listed there, including controlnet. Alternatively, you can add your own models by just duplicating the code.
I used to really love a previous colab that you made because it was easy to run SDXL + the refiner in one go.
Is it possible on this version?
I have so many different notebooks. Which exactly do you mean?
i'm a happy subscriber now. keep up the good work bro. 😀
wow, thank you very much 🩶
Hi, is there a way to stop the last cell from disconnecting? It works fine for roughly 20 minutes and then the server disconnects with the cell in colab showing a tick to say it's done, I then have to start the whole process again if I want to try it again and it's really tedious. I'm seeing very few comments with the same issue I have.
I've tried getting ChatGPT to help but it didn't get a solution. Please help me if you can and also, thanks for this notebook because it does work well for 20 minutes at least.
Update: it doesn't work at all now with the 2nd and last cell completing quickly, skipping the server generation completely.
At the moment, it seems that the notebook is still not blocked and is functioning normally. It might be due to regional restrictions; you could try using a different account with a VPN
Hello, thank you very much for this colab, hope it's not gonna be patch soon, i have a question, can we use upscalers? It don't seem we can
I'm glad that it's helpful! Of course, you can. I even showed it in the video.
And what about uploading other upscaler and change the denoising and stuff? @@marat_ai
Hi marat, thanks for the guide! Do you know if Invoke supports extensions like mov2mov?
I don't thing it does, only manually. Check custom workflows
@@marat_ai I'll try that out, thanks again
Thank you for the video, I have a problem, I cant enable controlnet, and the "+" symbol doesnt work. How can I fix this issue. And do you have any advice to make similar to illusion diffisuion images at the hugging face website? Thank you
You're welcome! To add a model, try editing the YAML file, as most of the models are already listed there, including controlnet. Alternatively, you can add your own models by just duplicating the code.
marat, a few problem. already change control net from false to True , but i cant add controlnet in the UI
I got problem here, I cannot run the Stable Diffusion
NameError Traceback (most recent call last)
in ()
7 from pathlib import Path
8 from typing import Union
----> 9 clear_output()
10
11
NameError: name 'clear_output' is not defined
Hey Marat, when I generating anime pics using Invoke AI the result is not that good compared with Automatic 1111
Is it because Invoke AI itself or because some models that you added by default?
Hi! Yes, It’s only because of the model, just try to choose the right one
Wait so I actually have to take the time redownload everything everything or run the code or whatever everytime I reopen google colab?
Does this Version of Invoke ai Have support of SDXL?
of course
Hello marat, thank you so much, unfortunately google collapse keeps crashing on my unstable diffusion -.- What can i do?
Although you preface this notebook by stating 'You don't need additional Google Drive storage because uploaded models are not stored on your Google Drive. After the session ends, all data will be deleted', is there anyway to connect my Google Drive regardless? I already have Loras, VAEs, Embeddings and Models downloaded onto my Google Drive for ease of access.
I've already made a version for Google Drive, available on my Patreon. I talked about it in the video
I got an error when I tried it today. Could you please review and fix it?
Can we use SDXL in this notebook? Pls reply
Sure but it's not so stable, you can also check out my latest video. There is a UI with SDXL under the hood.
thank you. i have watched.@@marat_ai
Hey how to get the controlnet? there is just one available (depth)
Can you fix the error thats not letting the models load nor generate images please 🙏🙏🙏🙏
Hi marat , your English is so good ❤
Wow, thanks so much! I'm actively working on it:)
Thank you very much for the very good information. I was able to install, start, and generate. So I was able to add the lora file and it was reflected, but I tried to download and use the important model from civitai, but after 4 hours of trial and error, it didn't work. If there is a site where I can find an easy way to do it, please let me know. I added the yaml but kept getting errors.
You're welcome! I added a cell for more convenient downloading of base models.
it tells me that the google collaboration colud is full, how can I delete everything inside?
Bro can you update the upscaler again too?. The image are blurry after upscale.🤔 can you add other upscaler mod?
how can I avoid disconnection?
can i using civitAi model and lora in that?
can we add extention for example like deform?
I didn't realise when signing up for this, I would have other people pictures in my drive, Is that normal?
Thanks for the subscription! As for your questions:
In the case of the Google Drive version, you only need to run steps 1 and 3, as all the necessary models have already been downloaded during the initial run.
It's very strange and definitely not normal. It's possible that there are some bugs with Google Drive. Can you provide additional information? Check the data of creations and ownership; it might clarify the situation. Maybe it's related to generated images...
it always disconnects after I use it for half an hour, and I have to repeat all the process again
do you have any solution?
Seems there is a problem brother... It doesn't want to load or import any models and lora
Good job bro
Deforum works are missed on UA-cam, I hope you're planning to bring a colab work for deforum.
Thanks for your simply-explained works
You are welcome! I'll see what I can do.
Whenever i install any new model through the Model Manager and it downloads it but when i change the model and press invoke the cell stops running the final thing it show is ^C
Days ago I found a youtube video that had a colab to generate images but the difference was that it did not have webui, by chance you have it?
I don't have one, but I could consider creating one, perhaps in the next few months for a special video series.
It would be easy to look for.
can i have the link for that video you are talking about?
@@mohammad815 I lost it, that's why I ask
I don't understand how to add new models to this google colab thing. Every time I think I'm doing it right, something breaks.
it is not working for me, the step 3 does not give me a link, what is wrong?
there's an error, no models loaded and unable to import model.
Ok, got a few problems. The generated images are weirdly low quality, with weird faces and soft blurriness. And then I am getting this error "Error while invoking:
a Tensor with 2 elements cannot be converted to Scalar". What do I do?
Experiment with steps, samplers, and models; there's very little chance of a problem in the invoke itself.
I wish the UI worked better on Mobile, any tips?
It was working 12 hours again but the last step is not working like annaa is saying. We have to take alook at it again
can you make a notebook for invoke version 3.6
This is not working at 2 step now
Can I download models from direct links?
How to install animatediff on the notebook? I've used a1111 so far and of course it's blocked
Can I achieve it by just writing a few lines of code?
Unfortunately, not yet
its work nice but when you put a link for a chekpoint collab just crash...
This is awesome! I just need some time to get used to the UI. Thanks anyway!
patchmatch suddenly not working (won't load/compile), any idea how to fix?
recomment this because I thought I found the solution, still didn't find it unfortunately
welp I am dumb and comment "exit()" that is used to restart the runtime and is needed for everything, idk why I comment it in the first place
can we add t2i adapter ?
hey got any ideas on how to download lora and embeddings from the invoke itself.
Sure, just watch the video tentatively
is there anyway to use my google drive space so it doesnt use as much of the jupiterlab space? like for models, loras, etc?
Yes sure, I've mentioned in the video, I have googledrive version in my patreon
@@marat_aibut I'm talking about jupiterlab
how to continue after load model? there is no tutor in video, is this new feature?
sorry for asking
i cant run step 3 after load other model from civit ai, but thanks i really helped by this video, my laptop doesnt have gpu and i use cpu and its need 2 hour for basic prompt, so thanks
I can no longer start the SD, the error is “Unable to connect to the accelerator (GPU) due to usage limits in Colab.” Can you help with it?
Do I understand correctly that you need to buy a subscription, and then you can use the service at any time?
@@snajdnrj In both yes and no, there is a daily quota of approximately 5 hours per day. You can just use another account or wait one day
thank you so much for your work and your answer!@@marat_ai
thank you for your work, watched it many times
Wow! Thank you so much!
@@marat_ai colab works fine, but sometimes got shutdown because System RAM full and barely use any GPU RAM. thanks for replying
@@ProjektRaven is it happening with ver 1 or ver2? I recently updated the notebook
@@marat_ai oh yah, i was using ver1 copied yesterday. and now im using ver2, the thing still happened high System RAM low GPU RAM
No masking or local redrawing functionality?
getting a error, it can not find the file in step 2 i need to edit to configure the models i download,
Not able to add model from civitai
Can you tell me how can i create deforum videos with it please 🚨🚨🚨🌹🚨🚨🌹🌹🚨🚨🚨🚨🚨🚨🚨🚨
Can add more models to the yaml file?
Of course, duplicate any needed model's codes (Lora, checkpoints, etc.) and just edit them. You can also save the file so that you don't have to edit the notebook every time; next time, just paste the whole code.
can it do batch img2img diffusion?
why i can't upscale the image? it says failed to queue graph
The RAMprovided by Google for free is only 12GB. It is easy to force the connection to be disconnected due to insufficient RAM. Is there a way to solve this problem in the free plan?
So, should we become a member of Google colab if we want to eliminate this problem?
Sometimes it's happening, most likely with SDXL. You can try restarting the session with the 'exit()' command, you can also find it in the notebook code, I commented it with #, just delete # sign
Not working. Interrupted after converting model to diffusers
The link is not found in the description
not using yet
but I guess it should run better than running on my 2gb vram potato
thanks
What graph do you have?
Hi bro instead of this invoke can you give the old notebook with sd1.5 inpainting version as you posted in first video
Oh my hero, your're trully life safer my brother!!
Your video is really great, so easy to follow. I have watch it many time to support you. I want to create a simple text to video generator by using sequence image as frame. Can you give me some guide or just some idea of how to do it?
Thanks for the support! You can use almost any video editing tool to extract frames, or you can even use online services. Then, you can manually or automatically (depends on UI) process them in SD.
@@marat_ai is it possible to generate each frame in SD automatically with just text prompt? Is there any tool that support it or similar to it yet?
not working anymore . Colab now detects it as a disallowed code
how to set clip skip 2 ?
Update it to invoke 3.3.0