How to install Stable Cascade for Automatic1111 & Forge.
HTML-код
- Опубликовано: 3 июн 2024
- Stable Cascade from Stability AI is here and it's a new model for Stable diffusion. Let's look at it and let's get it installed for you all.
stability.ai/news/introducing...
huggingface.co/spaces/multimo...
github.com/blue-pen5805/sdweb...
Patreon guide for subscribers: / stable-cascade-98767380
Prompt styles for Stable diffusion Automatic1111, Forge, ComfyUI & Vlad/SD.Next: / sebs-hilis-79649068
ComfyUI workflow for 1.5 models: / comfyui-1-5-86145057
ComfyUI Workflow for SDXL: / comfyui-workflow-86104919
Get early access to videos and help me, support me on Patreon / sebastiankamph
Chat with me in our community discord: / discord
Stable Diffusion for Beginners Playlist • Stable Diffusion Begin...
My Weekly AI Art Challenges • Let's AI Paint - Weekl...
My Stable diffusion workflow to Perfect Images • Revealing my Workflow ...
ControlNet tutorial and install guide • NEW ControlNet for Sta...
Famous Scenes Remade by ControlNet AI • Famous Scenes Remade b... 
Хобби
Patreon guide for subscribers www.patreon.com/posts/stable-cascade-98767380
👋
What are the minimum system requirements?
I am getting this error AttributeError: module diffusers has no attribute StableCascadeUNet do you happen to know why? Thank you
Same. Pretty disappointing after three installs.
same
same thing here, i've tried editing files, using a fresh install of forge, different versions of diffusers.. nothing works. at a certain point it moved onto a different error about AttributeError: 'generator' object has no attribute 'image_embeddings'. meh
I am dying for Version 1.0 of Stable Cascade.
Not only because it'll be better but because of the community models and all the Loras we can make with a model that actually understands the prompts.
Just noting that Stable Cascade as it is now is aimed at 16 gig vram, there's ways to get it working even on 8 gig, but a single image takes 5 min. We'll see how's this model doing in a month or two.
Edito: The models released after the initial release are much faster. Getting images in 26 seconds in Comfy UI with a 3060ti 8gig.
I am running Stable Cascade in ComfyUI on an 8GB RTX3060TI, and it takes around 20 - 30 seconds to get resolutions between 1920x1152, and 1840x3072. It wouldn't surprise me if A1111 took 5 minutes, but ComfyUI is very fast. Will be trying Cascade in Forge next.
5 min for just the first image or all? because the first time you generate something it downloads models etc
@@Zibbibbo 5mins for each image after the initial load. The new two model solution for comfyui is slightly slower (seconds) than the 4 model version I was using yesterday. I can still generate 3072 X 1840 images in under 45 seconds using comfyui. Forge takes at least ten times as long for 1024 X 1024 image. Comfy also let's you play with the compression, which can speed things up, or let you make larger images.
@@guns1inger I guess I may try in Comfy, then : )
@@Zibbibbo As of today, getting images in Comfy UI in 26 seconds with the models they released after the first release. I was using the first released models and the first nodes for them. With those, the minimum for an image was 3 minutes.
I'm thinking about cancelling my chatgpt pro account and joining your patreon as I find stable diffusion so much more satisfying than the big tech generators
I mean chatgpt is fun and all, but there's a reason I've only done a few videos on it, and 99%+ on AI art 😅
The big tech models like bing were amazing for a brief moment but they destroyed them because they did their job too well. It’s a shame that they’re holding back such amazing tech.
I installed it using the 'install from url' tab and it shows up in the extensions list now, however there is no tab, any ideas?
I have same problem :(
Same
Hi, did you fix it somehow?
@@andresanders1132 No, if I manage to fix it, I will post i comment here.
Tried to install on my colab pro account. Same problem. Colab reported some errors when reloading the ui.
Having a bit of trouble. The extension is installed, but not showing up on the tab.
Same here, did you manage to fix this then?
@@alokdesaimusic no
This is awesome, thanks. So far Stable Cascade is a game changer for me. I'm rarely using negative prompts at all and very few positive prompt words to get what I want. Crazy how fast it is for the results as well.
Great to hear! 🌟
Hi, I got the stable cascade tab but it gives me this error while trying rendering. Any solution? module diffusers has no attribute StableCascadeUNet
Also on their github page there's written please have someone redo this extension... is it down?
also having this error, any updates on it?
@@JackedOnMtnDew Same
Great videos man, such a unique and chill vibe! I appreciate you.
Glad you enjoy it! Welcome aboard 😁
do you know how to change the stable cascade model? I see civitai has like 30GB models, curious to see if they're better quality
Speaking of Gnomes - how does a Gnome eat?
Gnome Gnome Gnome (nom nom nom)
Thank you so much. It was for me today to solve it on my machine. Perfect timing! peace
Glad it helped you!
Can we use loras?
I am getting some crashes during the install of the endless requirements. How's your virtual memory settings?
Hi Sebastian, I tried to install the estension and it seems installed but it doesn't appear in the A1111 tabs. It appear only on the list of the estensions. What can I do?
Do you have any idea where to place the upscalers in forge directories?
\stable-diffusion-webui-forge\models\ESRGAN
@@ClashMafia Its really strange, Thats the same path of the original webui and I'm sure I tried it first but this time it worked. Must have done something wrong. Thanks for the help
Using the one click installer for Forge, when I installed this I get this error:
*** Error loading script: easy_stablecascade_diffusers.py
Traceback (most recent call last):
File "C:\Stable Diffusion Forge\webui\modules\scripts.py", line 544, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\Stable Diffusion Forge\webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "", line 883, in exec_module
File "", line 241, in _call_with_frames_removed
File "C:\Stable Diffusion Forge\webui\extensions\sdweb-easy-stablecascade-diffusers\scripts\easy_stablecascade_diffusers.py", line 5, in
from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline
ImportError: cannot import name 'StableCascadeDecoderPipeline' from 'diffusers' (C:\Stable Diffusion Forge\system\python\lib\site-packages\diffusers\__init__.py)
Loading it again using the non-one click method in your other video worked perfectly. You do have to uninstall python though first.
I keep being amazed at Ai art generation, so freakin cool
Right!? :D
Where did it put all models ? I can't see them.
What am I missing here? it's taking a hella long time to generate in A1111. Are there specific models I should use to generate?
I used the 1-click installer for Forge so if there's issues I'll pass. I have it through Pinokio anyway, so it's not like I'm missing out.
Maybe I'll try on my A1111 install first.
Can you achieve something similar to Heygen? Mainly lip syncing?
I've tried using others but without success
Hi friend, I installed Stable Cascade in my Automatic1111, but the Cascade tab does not appear in the interface, what could be the problem?
RuntimeError: Failed to import diffusers.pipelines.stable_cascade.pipeline_stable_cascade because of the following error (look up to see its traceback):
cannot import name 'List' from 'typing_extensions' (C:\ai\stable-diffusion-webui\venv\lib\site-packages\typing_extensions.py)
Is it normal 1500x1500 render on a 4080 takes about 90 seconds? Where are models? I saw during the installation of the extension that long files were downloaded, I can't find them in the A1111 folder. Thanks for help.
Normal
I have a 4070. Same issues for me. Its soooooo slow. Also pretty censored. That will change over time but the slowness is deal breaker for me.
huggingface cache folder, by default it is your home directory/.cache
Are we able to use controlnet with this at all yet ?
Hello, I'm one of those people for whom it didn't go well (30min Oo), despite my 4060ti, but on Comfy it's going better for me.
Hello Sebastian ! I have a little question, i am using both A1111 and Forge but I have the feeling that 1111 has higher quality images especially when using controlnet, or is it just a feeling ?
I'm not sure how the backend handles that differently tbh. I don't see why a1111 would give you higher quality, but I also can't say for sure. My guesstimate is that it's a feeling.
@@sebastiankamph Yes you are probably right, i just made some tests with the same values and prompts and A1111 gave me better result with the controlnet, perhaps it was just random, i didn't do much tests yet to compare so, i wanted to ask you your thoughts :) thank you ! And thanks a lot for your awesome content !
Thank you for all your amazing contents ... I just saw your video and based on the introduction page model c is the base model and model b is the refiner I guess? and model c is the vae , so I was wondering how can I use it in automatic1111 , I just saw that you had some kind of merge model with sdxl (based on the name of the model in your video)? Can you help me how to merge it? Thanks
Hey, it's not using that main model, but is downloading a separate model that it runs from inside the extension. You can't select it currently.
Holy shit. That is incredible!
Installed Forge and then the Stable Cascade tab, when trying to generate I get "Runtime Error: cutlassF" no kernel found to launch!" any thoughts?
Same boat. Commenting in case someone finds a solution.
sorry but where are the images being stored?
I could not find them in the directory
i've installed the extension and am able to select the tab but it never actually generates an image, despite the loading animation taking place. anyone else have this problem? i am running automatic1111 1.8
So, its just an extension? You can install it into any instance of A1111? How do you get the models or install them?
+1
Ok those hands it generated were unreal good
It still messed some up, but many turned out very cool!
I did the install but it's not showing in Auto1111 even after restarting it. What can I do to fix?
Could you please make a video on the newly released, X Adapter?
Hi, where are models stored? In case, I want them delete. Tnx
which models can be used with this? and which one are you using in the video??
It's only using stable cascade main model.
Instaled cascade on both Forge and A1111 and when I hit generate its just processing for ever and doesnt generate anything. Anybody else getting this problem or know a fix?
hi i tried to install it in forge just like you did but i have no tab that says stable cascade but it says its installed
Maybe someone can help me. i installed cascade for A1111 but if a try to generate a image i get this message: RuntimeError: cutlassF: no kernel found to launch! and several other lines.
Does the model you have loaded as a checkpoint matter? I don't think you mentioned it.
got error: ModuleNotFoundError: No module named '_socket'
while installing the extension using given link.
installed forge using 1-click installer.
How did u fix??
I get the fastest speed with Stable Cascade using Pinokio ~13 seconds vs. 21 seconds using the extension in Forge (4060 ti 16gb). Auto1111 get's confused and try to run multiple checkpoints i think because it takes over 5 minutes for one image. Normal SDXL generation normally takes me 9 seconds to complete on Forge/Comfyui and 10 seconds on Auto1111. 6-7 seconds with tensorrt in Auto1111. I can't get Forge to work with Tensorrt properly, not even when i copy the 'Unet-onnx' & 'Unet-trt' folders over from Auto1111.
For those of you having problems getting it to run in forge that was installed via 1 click installer. So when I run it from the main directory using run.bat, then the stable cascade tab will not show up. However if I run webui-user.bat from the webui directory instead of through the run.bat, it shows up correctly in forge without any issues.
i have installed All my WebUI´s through Stability Matrix i tried to install Stable Cascade in Forge but all im getting Error: No Kernel Found
HI, I have 16 vram but after pressing Generate it takes forever for Cascade to generate. Why?
Cascade is really slow tbh. Is it worth it? 50/50
🎯 Key Takeaways for quick navigation:
00:00 🌟 *Introduction to Stable Cascade and its Installation Process*
- Introduction to Stable Cascade and demonstration of its integration with Automatic1111 and Forge.
- Explains the ease of installation and the benefits of Stable Cascade, including faster performance and improved prompting.
01:20 🛠️ *Technical Overview of Stable Cascade*
- Overview of Stable Cascade's technical features and benefits.
- Highlights the model's efficiency, high-resolution capability, and comparison with other models for speed and quality.
02:58 📲 *Installation Guide for Stable Cascade*
- Step-by-step guide on installing Stable Cascade via URL.
- Mention of Patreon support for additional resources and text guide.
03:53 🔧 *Troubleshooting Installation Issues*
- Solutions for common installation issues, specifically for users of Forge and Automatic1111.
- Tips for manual installation to resolve problems.
04:34 🎨 *Demonstrating Stable Cascade's Capabilities*
- Showcases the model's ability to generate images based on varied prompts, including fantasy, Sci-Fi, and Manga styles.
- Examples illustrate the versatility and quality of Stable Cascade's output.
06:00 🌱 *Advanced Prompting and Fine-Tuning Results*
- Discussion on advanced prompting techniques for more complex scenes.
- Comparison of results with previous models and the improvement in capturing detailed scene elements.
07:34 💡 *Highlighting High-Resolution Capabilities*
- Emphasizes the ability to generate high-resolution images natively.
- Showcases the quality of detail that can be achieved, even on high-end hardware configurations.
08:44 🏁 *Conclusion and Patreon Acknowledgment*
- Concludes the demonstration of Stable Cascade's features and capabilities.
- Thanks Patreon supporters and teases future content.
Made with HARPA AI
So no way to use it with batch sequencing and controlnet yer, right?
Not that I've seen, but maybe in some fancy Comfy workflow?
Got regular Automatic1111, installed the cascade, it does show up active in extensions, however I can't see it on the 'main' screen of the SD interface, weird D: Maybe it has something to do with set model? Mine is 1.5, not SDXL
I've got the same problem, but I've got SDXL loaded (it also doesn't show up with 1.5).
Same with me.
Same here@@mharris7380
Same for me.
This is mentioned in the video. You used the one click installer and it doesn't work with that for Forge and A1111. You need to do a manual install then it will. Or use Stable Cascade by another method like Pinokio or Comfy.
Did I miss how instal model or it's not in the video ?
Extension autodownloads
Bro could have mentioned that on first render it TAKES FOREVER to download everything and not to panic that installation failed! :D So much to download, its nuts!
hi can you help me to solve this error "module diffuser has no attribute StableCascadeUnet. i installed cascade install in stable diffusion but i got this error after installing of all model on window 11
I have the same error. A few days ago everything was still working.
Also got the same issue, was trying to fresh installation of forge and doesn't help at all.
ModuleNotFoundError: No module named '_socket" Thoughts?
Didnt work on macOS, complained of pyTorch, which is surprising as I can A1111 without any problems. Using Pinokio, we can start up at least but it gives memory error, on a 24GB machine
It takes about 160 seconds per 1024x1024 image for me with a 3070 GPU. It's much faster with 512x512 but that seems to only create a centre crop of a larger image on first try.
i installed this. got a 12gb GFX card. 48gb RAM.
prompted a simple thing. thing is still rendering 200s. taking FOREVER to prompt
I don't know about it being faster. Super slow on the 4080 super. No sure what could be the issue. Wonder if its a bottleneck. I have a 4080 super that is connected to a riser (maybe this riser is the issue). Running with a 5800x. 64gb or ram.
I installed forge manually but can't launch it, any ideas how to get it running
What's the issue?
@@sebastiankamph it's saying it's missing a luanch kernel
@@springheeledjackofthegurdi2117 same here. Might be related to how much GPU ram is available. I found this "fix" but nobody explains where to actually put this code:
torch.backends.cuda.enable_mem_efficient_sdp(False)
torch.backends.cuda.enable_flash_sdp(False)
maybe someone here knows.
@@sebastiankamph Im getting the same kernel error
RuntimeError: cutlassF: no kernel found to launch!
is this only for 16gb vram?
People claim to be running it on 8, but my understanding is it's meant for 16+
Nice one mate! I told ya Forge was fast lol
You got that right! Vrooom
You said it's uncanny where ironically the ghibli cat can be achieved with some kind of canny controlnet function. Though not intended, I thought it was funny.
gracias
Rather than generating new images, I inpaint existing photos to change one or another detail.
Thanks
I'm getting error on my Macbook M2 Pro Max 32G
Here are my results in a low spec system(16gb RAM/3060 12gb):
It works, but forget about lower res images, the precise txt-to-img results only work consistently when using 1024x1024+
It is very slow, it is loading the model from the SSD every time I try to generate images, only after that is actually using RAM and VRAM.
A 1024x1024 image is taking around 80s, as comparison, using dreamshapperXL Turbo I can generate similar quality images under 8s.
Sooooo you state it's an extension but of what program????
For automatic1111 and Forge
Thanks@@sebastiankamph
Not works (A1111 here) no tab "Stable Cascade". And why is there no mention whatsoever of downloading the model?
Model is autodownloaded. Update your a1111
@@sebastiankamph Thx. Already updated. I don't know what it is. In any case, I tried with Pinokio and my 4070 it's not enough with its 12 GB of VRAM. Maybe with future models with smaller Stage dimensions. 🍀
Ok maybe I'm stupid but why would I use cascade when forge using the same model and size is 5x faster to render, what makes it unique??
It's a different model, it has better prompt understanding and claims some improvements over sdxl. But it's more of a lateral step, like dall-e 3.
can it work with rtx 3070 8gb? thanks
Don't know but if it does it'll be slow. It takes several minutes (4-5) with 12Gb VRAM.
Should work, and should be faster than that if you're not doing huge resolutions like 2048x2048.
i have a 3080 (10gbvram) and found a work around. cascade wouldn't even work until i added --medvram to the set COMMANDLINE_ARGS= line in the webui-user.bat launcher
Great tip!
Why is always the same, this doesn´t work in the AMD stable diffusion option? I´m so tired of this. I installed and don´t get any tab of cascade in the options.
Did I understand it right that stable cascade has a non-commercial license, so we can't use it to create sellable Art, right?
That seems to be the case
@@sebastiankamph so sad 🥺 that is the case for SVD and for Cascade and also for turbo...
They have no real control over something like that. What they don't want is a bunch of online apps etc using their software to generate people.
You're using it as a tool, not selling it as a tool, you'll be fine.
@@tbird81 so do you think the results generated with it would be okay to sell? 🤔
@@sirmeon1231 I don't see why not. People use pirated copies of Photoshop to create and sell art.
Morally there's nothing wrong with what you're doing. Who knows/cares about legally. Anyway, at the moment it's early days. Go for it.
You should have shown the differences between StableCascade and the regular A1111 in the video
no loner works even with manual install. huggingface changed diffusers i think
It can do 4K (3840x2160) for landscapes.
Oooh, very nice. How are your results?
@@sebastiankamph My desktop wallpaper is beautiful. :-) 4K monitor. It can also do an aquarium and nebulas / starry skies. There might be other subjects that work at 4K, but those are the only ones I know of.
With max compression (64) 3072x2048 works as well as 1024x1024. High CFG on stage C for complexity / details. The sampler / schedule you use will make some subjects impossible. I have img2img working in a python script, but no support in the UIs yet.
Does it work for 6gb vram?
no
Only on ComfyUI... But it's slow generating images, like 5 minutes each...
where are the models being installed?
It downloaded 8+ GB, didn't work but I want to know where the folder is
Edit: nevermind, already found it: C:\Users\**********\.cache\huggingface\hub\models--stabilityai--stable-cascade-prior\snapshots
If You have a less powerful Nvidia GPU, try the stand alone version stable cascade easy (you cant post links here?). With my RTX 3060 12 GB a 2048x2048 image takes 73 seconds to generate and 1024x1024 image 26 seconds.
For some reason, it is giving me fuzzy images :(
yah that is a whole lotta don't work in my A1111
All right, I'm not just here for the little-gnome fact - bring on the Stable Cascade.
I got you covered
Can it work on MX450 2GB graphics card ??
Yes, but it will be finished next year.
hmmmmmm.......... jury is still out on this one
Well maybe other RUclipsrs were faster, but other youtubers are not Seb and its not a good watch without a dad joke 😂
You got that right! 😅
30GB. Good heavens
not working for me, taking ages to download models (45mn still at 20% for one of them) I'm removing the cascade extension but does anyone know where the already downloaded models are stored ? there are no models in the C:\Users\****\pinokio\api\forge.git\app\extensions\sdweb-easy-stablecascade-diffusers
thanks !
Stable cascade is generating garbled images for me, maybe it's the 12 GB gram limitation 🤔
*** Error loading script: easy_stablecascade_diffusers.py
Traceback (most recent call last):
File "C:\Users\xx6\Documents\A1111 Web UI Autoinstaller\stable-diffusion-webui\modules\scripts.py", line 469, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\Users\xx6\Documents\A1111 Web UI Autoinstaller\stable-diffusion-webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "", line 883, in exec_module
File "", line 241, in _call_with_frames_removed
File "C:\Users\xx6\Documents\A1111 Web UI Autoinstaller\stable-diffusion-webui\extensions\sdweb-easy-stablecascade-diffusers\scripts\easy_stablecascade_diffusers.py", line 5, in
from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline
ImportError: cannot import name 'StableCascadeDecoderPipeline' from 'diffusers' (C:\Users\xx6\Documents\A1111 Web UI Autoinstaller\stable-diffusion-webui\venv\lib\site-packages\diffusers\__init__.py)
Yeah I'm got the same thing. To fix it... uninstall python from add/remove programs. Then follow the manual method of installing git/python/forge again here in his other video: ruclips.net/video/mHvnHA04sbc/видео.html&ab_channel=SebastianKamph Worked just fine for me.
We'll see uf this model becomes any useful for anything in the future, I'm skeptical. There was also a hype around SD 2.0, 2.1 and so on and the versions turn out to be massive fail. Even the SDXL isn't as good as useful as SD 1.5 imo. I'll check it in month or two if the SC is any good. Hype is for children.
* Error loading script: easy_stablecascade_diffusers.py
Traceback (most recent call last):
File "C:\Stable-Diffusion\Forge\webui\modules\scripts.py", line 544, in load_scripts
script_module = script_loading.load_module(scriptfile.path)
File "C:\Stable-Diffusion\Forge\webui\modules\script_loading.py", line 10, in load_module
module_spec.loader.exec_module(module)
File "", line 883, in exec_module
File "", line 241, in _call_with_frames_removed
File "C:\Stable-Diffusion\Forge\webui\extensions\sdweb-easy-stablecascade-diffusers\scripts\easy_stablecascade_diffusers.py", line 5, in
from diffusers import StableCascadeDecoderPipeline, StableCascadePriorPipeline
ImportError: cannot import name 'StableCascadeDecoderPipeline' from 'diffusers' (C:\Stable-Diffusion\Forge\system\python\lib\site-packages\diffusers\__init__.py)
how to solve this error ://
what i've done. in the stablecascade extension folder there is the install. py file. inside you can find the diffuser link "git+..." (i can't put it here.. youtube is strict)
now as the python version of the one-click-installation seems broken i create another python environnement (3.10.6) same version as the one forge use then under that environment do "pip wheel " (better do that in dumb folder as it will create WHL files for diffuser dependencies too..
now that you got a WHL file for diffusers do 'pip install --no-deps " (make sure it use the forge's python this time.. you can do C:/path/to/specific/python.exe -m pip install --no-deps blabla)
by the way inside the webui folder remove "diffuser==x.x.x" of any "requirements_x.txt" otherwise forge will keep installing the wrong diffuser at the ui's start and you will keep getting this error
Same
You created a new environment for all of automatic1111 or just for that extension? These instructions are very, very hard to understand.
@@UnresolvedExternal you can try to install by using forge python installation but i got an error when i did that
do "forge-webui\system\python\python.exe -m pip install "
8Gb VRAM takes FOREEEEEEEVER. This is for rich peoples.
никто не генерирует SDXL модель в 50 шагов! это читерский график, у всех в серднем 20-30
Ah poo, not working with AMD yet. :(
Dunno... For now Stable Cascade looks like over-complicated model with several extra "moving parts" which produces same plastic or airbrushed people as SDXL... Far from photorealism.
but u made 1 click installer. for patreon only... nasty
My dear, thank you for the video. Please go away with the background music it is loud enough to be disturbing and distracting from what you are trying to present. Or at least lower the volume of the background music. I actually stopped the video at 2:36 because I was unable to follow what you were saying and my eyes were squinting trying to focus. Thanks!
Thank you for the feedback!