WARNING: this software killed my gpu. gtx 1070. been using it for 6 years with no issues. used this to render ~100 images and it's completely dead. so uh, maybe don't run this with your card overclocked? idk, weird
Can anyone with a 3090 Ti confirm their highest resolutions possible as a result? Looks like I can max out at 768 by 512 on my 2080 Ti. Any examples would be appreciated.
I have a 3090 TI and I was actually going to ask about the same thing in the comments. For some reason it won't use more than like 13 gigs of vram out of my 24, and if I make the resolution bigger than 512 by 768 or something around that it gives the "ran out of vram error"
Like it says that it tied to allocate 3.7 more but it also clearly says I have 7.9 available.
But honestly 512 by 768 is good enough for me just playing around I don't really mind that much.
Thanks for confirming. Shortly after I ended up reading a ton of articles and it appears it wouldn't matter much anyways, as they were trained at 512x512 resolution to begin with, which makes sense as to limits and images tending to overlap or duplicate when you set the output higher. And you're right, 768x512 is really perfectly fine, I was just curious but knowing more now realize that there wouldn't be much benefit of a 3090 Ti over a 3080 Ti... at least, yet.
I can do 1080x1080 on the updated version of the GUI using my 8GB 2070 Super, and still have enough VRAM left over to fiddle around in Blender while the GUI does its work.
It's kinda annoying that they're paywalling such big updates, I hope they plan to update the Itch version soon so people don't have to pay $10 a month for access to a bunch of different services they probably wont use, just to get the bugfixes, content, and optimization of the newer versions of the GUI.
Doesn't really work. Worked fine once, then now every time I try to render something it just keeps saying CUDA out of memory, and this persists even after a hard restart. I have a 2080ti with 11G of ram, and the program can't even allocate 3GB to render a file. It seems the program is allocating memory that it never releases after its done, and as such each subsequent load hogs up GPU memory.
I don't think the program, or the Stable Diffusion implementation, is ready for prime time yet. The limit of 512x512 is about as far as I can render, which is also really low resolution - and even with Gigapixle the quality is subpar to other programs like Midjourney.
Sorry to hear it's not working for you. I've got a RTX 2070 works great (8GB of VRAM). Maybe Midjourney is a better option for you if you don't mind using Discord.
I love this tool so much that I decided to become a Patron! It's so useful to have a separate txt file with the settings used for each generated image (as opposed to have it in the file name of the image like the Dreamstudio). I'm just downloaded GUI version 0.41 and I was happily surprised to find out I can generate larger images, samples per prompt works and there is an upscaler as well as inpainting!!! (although I haven’t figure out how it works)
Anyway, I thought I’d share this since more people might decide to support the development of this GUI as it’s not advertised anywhere what’s in the Patreon version (or at least I didn’t know, I just thought I was giving money to say thank you!).
Are there plans to add GPEN or GFPGAN to fix faces? That would be amazing, along with the ability to save the output path!
Is there a reason why I'm getting randomly images generated without any input from me ? I generate an image then it runs again and outputs something else
An example of info from one of the text files associated with the rogue image.
The latest Patreon version does not support the 16xx series yet. And you need the 9$ tier for the downloads. Might want to wait until half precision can be turned of though
I tried it on my with a OptiPlex 990 / 16 GB RAM with GeForce GT 710 but I get this error:
File "torch\nn\modules\module.py", line 925, in convert return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking) RuntimeError: CUDA out of memory. Tried to allocate 44.00 MiB (GPU 0; 2.00 GiB total capacity; 1.57 GiB already allocated; 20.93 MiB free; 1.63 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Anyone know if there anything I can do to get it to run on this machine?
Thanks for your advice. When you say turn down resolution, what do you mean? I dropped the Resolution down to 64x64 but I got the same error. Currently I have 16 GB RAM. How much RAM is the minimum requirement? Thanks again!
2 GB, so I see I need 4 GB. Hm. Someone recently mentioned that I might be able to create a virtual machine and up the VRAM using it, but I'm not sure if that would work given the NVIDIA CUDA requirements. Hmm... Maybe I need to upgrade my hardware after all.
Hey GRisk, amazing tool! I'm also running a 1660 Ti GPU, so no results for me right now. Would it be too much to ask to add a feature to disable the half-precision mode that's at the hearth of why it's not running on 1600 series GPUs?
it works on my old 1080, and I think I saw 1070 working somewhere. Other than these you would want an RTX 20xx or RTX30xx series All Nvidia, no AMD support
Is there any way to manually update the libraries from official sources so that this app uses them? As written below, a new version is out, want to try it
i know its the same model lol. he literally updated this post to include his patreon. paywall blocking for basic GUI features is disgusting and malicious, especially when he has 500 supporters.
I tried at windows, the output are all black images, I get this prompt error while the gui is loading:
‘’’
torchvision\io\image.py:13: UserWarning: Failed to load image Python extension:
torch_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: <function _DenseLayer.forward at 0x000001F6781C65E0>.
warnings.warn(f”Unable to retrieve source for @torch.jit._overload function: {func}.”)
torch_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: <function _DenseLayer.forward at 0x000001F6781C6820>.
warnings.warn(f”Unable to retrieve source for @torch.jit._overload function: {func}.”)
‘’’
so I believe that maybe I have the wrong python installed or something like this? I’ve just unziped the files and executed the .exe as manager so in theory, it should work.
Could be, but im pretty sure its not. There is a setting to redner a second image, that could be it. Furthermore i am not having this issue, and i cant find any other evidence.
pretty sure you have a line break after your prompt, and it is generating an "empty" image because of the empty line... happens often when you copy your prompt from somewhere etc, or press "enter" at the end
I've had that happen a few times, it happens when you accidentally leave an empty line after the end of your prompts, like if you hit ENTER one too many times, or left a line after deleting a bunch of prompts.
I have a Tesla m60, which works like an SLI card. Can we add functions like;
Use all GPUs or, Check with a marker with GPU (0), GPU (1), etc? as you can see, GPU 1 is not using it's resources, With that maybe I can reach full of 16 GB's of Vmemory
There is no NSFW filter here, though I'd assume the Stable DIffusion model (out of GRisk's control) probably isn't trained on anything more lewd than nudity.
erry girl longshot, blonde curly hair, flowers crown, hyper realistic, pale skin, 8k, extreme detail, detailed drawing, trending artstation, hd, fantasy, d&d, realistic studio lighting, by alphonse mucha, greg rutkowski, sharp focus, elegant with graffiti background model shoot, apocalypse took like 10 mins do not recommend seed 3314742242 steps 50
Getting the following error: RuntimeError: CUDA out of memory. Tried to allocate 30.00 MiB (GPU 0; 2.00 GiB total capacity; 1.59 GiB already allocated; 17.26 MiB free; 1.62 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
My VGA card is a GeForce GT730, can i do anything with it?
Hi the only way to do this would be to try and generate the image at a smaller size. the vram recommendation is 4gb min and 6gb to generate images at the standard size of 512 by 512
This program requires Nvidia cards so it will never run on a Mac and AMD doesn't care about artificial intelligence (unfortunately for them) so cry yourself a river.
← Return to tool
Comments
Log in with itch.io to leave a comment.
Every time its render like this black image
anyone knows how to fix this
what computer do you have? and settings on the gui
asus tuf fx505dt
GTX 1650 4GB VRAM
Settings are default on the gui
Kinda sad that I can't use it at the moment due to having a 1660, but if it gets patched and fixed in the future I'll be hella happy to use it.
WARNING: this software killed my gpu. gtx 1070. been using it for 6 years with no issues. used this to render ~100 images and it's completely dead. so uh, maybe don't run this with your card overclocked? idk, weird
Hey, is there an AMD equivalent anywhere?
Neural networks, in principle, cannot work with AMD due to the lack of cuda.
Understood, thanks
No it will work with AMD. You will have to build it with pytorch thats based on AMD Radeon open compute (RocM).
Can you Make short video or instructable for that?
It will be really useful for a noob like me.
The image generation model is made by Stability Ai, it is open source. So obviously its free.
Got these images using prompts like - "old stone castle ruins in the rain with a river running alongside it and ivy growing on it"
Works great on my humble 1060 with 6GB
Generated the following image with user Ramonisai's prompt text
"
beautiful girl longshot, red hair, flower crown, hyper realistic, pale skin, 4k, extreme detail, detailed drawing, trending artstation, hd, fantasy, d&d, realistic lighting, by alphonse mucha, greg rutkowski, sharp focus, elegant"
how much time when generating this picture?
1 minute 20 seconds
Can anyone with a 3090 Ti confirm their highest resolutions possible as a result? Looks like I can max out at 768 by 512 on my 2080 Ti. Any examples would be appreciated.
I have a 3090 TI and I was actually going to ask about the same thing in the comments. For some reason it won't use more than like 13 gigs of vram out of my 24, and if I make the resolution bigger than 512 by 768 or something around that it gives the "ran out of vram error"
Like it says that it tied to allocate 3.7 more but it also clearly says I have 7.9 available.
But honestly 512 by 768 is good enough for me just playing around I don't really mind that much.
Thanks for confirming. Shortly after I ended up reading a ton of articles and it appears it wouldn't matter much anyways, as they were trained at 512x512 resolution to begin with, which makes sense as to limits and images tending to overlap or duplicate when you set the output higher. And you're right, 768x512 is really perfectly fine, I was just curious but knowing more now realize that there wouldn't be much benefit of a 3090 Ti over a 3080 Ti... at least, yet.
I can do 1080x1080 on the updated version of the GUI using my 8GB 2070 Super, and still have enough VRAM left over to fiddle around in Blender while the GUI does its work.
It's kinda annoying that they're paywalling such big updates, I hope they plan to update the Itch version soon so people don't have to pay $10 a month for access to a bunch of different services they probably wont use, just to get the bugfixes, content, and optimization of the newer versions of the GUI.
if this works for me was wodering if these are real random generations and if so i could use em in my own programs right?
McAfee has identified "Stable Diffusion
GRisk GUI.exe" as containing a virus. Is anyone else getting this?
who tf uses mcafee :skull:
I guess people that buy a new computer that comes with a subscription! XD
Doesn't really work. Worked fine once, then now every time I try to render something it just keeps saying CUDA out of memory, and this persists even after a hard restart. I have a 2080ti with 11G of ram, and the program can't even allocate 3GB to render a file. It seems the program is allocating memory that it never releases after its done, and as such each subsequent load hogs up GPU memory.
I don't think the program, or the Stable Diffusion implementation, is ready for prime time yet. The limit of 512x512 is about as far as I can render, which is also really low resolution - and even with Gigapixle the quality is subpar to other programs like Midjourney.
Sorry to hear it's not working for you. I've got a RTX 2070 works great (8GB of VRAM). Maybe Midjourney is a better option for you if you don't mind using Discord.
Random stuff I've generated.
Some really nice Ai images... well done... Especially loving the battleships at sea... gorgeous
Strange, I also have a 2080ti but i'm able to allocate pretty much all of my VRAM. I'm able to get away with 576x576 before maxing out.
I can do up to 1024 x 1024 with my 2070 (with Patron GUI mind you)
Holy shit i'm buying
I love this tool so much that I decided to become a Patron! It's so useful to have a separate txt file with the settings used for each generated image (as opposed to have it in the file name of the image like the Dreamstudio). I'm just downloaded GUI version 0.41 and I was happily surprised to find out I can generate larger images, samples per prompt works and there is an upscaler as well as inpainting!!! (although I haven’t figure out how it works)
Anyway, I thought I’d share this since more people might decide to support the development of this GUI as it’s not advertised anywhere what’s in the Patreon version (or at least I didn’t know, I just thought I was giving money to say thank you!).
Are there plans to add GPEN or GFPGAN to fix faces? That would be amazing, along with the ability to save the output path!
Is there a reason why I'm getting randomly images generated without any input from me ? I generate an image then it runs again and outputs something else
An example of info from one of the text files associated with the rogue image.
{'text': '', 'folder': '.\\results', 'resX': 512, 'resY': 512, 'half': 1, 'seed': 3314742242, 'origin': None, 'origin_W': None, 'steps': 100, 'vscale': 8.0, 'samples': 22}
That happens when there are empty lines in your text field. It doesn't know what to do with it, so it generates random stuff
Does the Patreon version support full precision so we can use the 1650 and 1660 cards? And which tier do we need to be on to access the downloads?
The latest Patreon version does not support the 16xx series yet. And you need the 9$ tier for the downloads. Might want to wait until half precision can be turned of though
Thanks, I guess I'll be sticking to the command line version.
I tried it on my with a OptiPlex 990 / 16 GB RAM with GeForce GT 710 but I get this error:
File "torch\nn\modules\module.py", line 925, in convert return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking) RuntimeError: CUDA out of memory. Tried to allocate 44.00 MiB (GPU 0; 2.00 GiB total capacity; 1.57 GiB already allocated; 20.93 MiB free; 1.63 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Anyone know if there anything I can do to get it to run on this machine?
either turn down resolution or get more ram
Thanks for your advice. When you say turn down resolution, what do you mean? I dropped the Resolution down to 64x64 but I got the same error. Currently I have 16 GB RAM. How much RAM is the minimum requirement? Thanks again!
i think its your vram actually, can you check how much vram you have?
2 GB, so I see I need 4 GB. Hm. Someone recently mentioned that I might be able to create a virtual machine and up the VRAM using it, but I'm not sure if that would work given the NVIDIA CUDA requirements. Hmm... Maybe I need to upgrade my hardware after all.
Hey GRisk, amazing tool!
I'm also running a 1660 Ti GPU, so no results for me right now.
Would it be too much to ask to add a feature to disable the half-precision mode that's at the hearth of why it's not running on 1600 series GPUs?
I have a NVIDIA GeForce 1650 card. All I get is a black png. Any idea how to solve this problem?
The models sadly don't work with 16xx series cards
Thank you so much for your answer! Do you know on which cards it does work?
it works on my old 1080, and I think I saw 1070 working somewhere. Other than these you would want an RTX 20xx or RTX30xx series
All Nvidia, no AMD support
Thank you so much!
Hi, SD 1.5 out! Is there any chance to upgrade?
SD 1.5 is not available for download as far as I know. Only available on dreamstudio yet.
can it work on a GTX 1070 with CUDA?
Works fine on my GTX 970 4Gb VRAM
I'm only getting black pngs. Can anyone tell me what's going on
Do you have a GTX 16xx series Card?
Mine is GTX 1650 and I also get black images
i'm the same comdition . maby check out the top information in this page will help you . graphic card have to surpport CUDA,1650 may not includded
Is there any way to manually update the libraries from official sources so that this app uses them? As written below, a new version is out, want to try it
Stable Diffusion 1.5 version is out! Curious to find out what has been improved. Hands hopefully among other things! Can't wait to try it.
Version 1.5 is not available for download yet afaik
Yes on DreamStudio web interface but should be available to download soon I heard.
it looks like hes just paywall blocking an updated version on patreon
the patreon version uses th same model as the itch one. just the GUI features are different
i know its the same model lol. he literally updated this post to include his patreon. paywall blocking for basic GUI features is disgusting and malicious, especially when he has 500 supporters.
I don't know if it's "malicious". He's done work, he wants to get compensated.
Don't have your hopes too high for v 1.5
Does it work on my NIVIDIA RTX 660 Graphics card?
I wonder why this program doesn't need that official 4g model, but a 1.6g one. How does it work?
What is a V scale?
CFG or Classifier Guidance Scale. Higher means it follows your text prompt more precisely. Not sure what the max is.
Stable diffusion is really powerfull.
I tried at windows, the output are all black images, I get this prompt error while the gui is loading:
‘’’ torchvision\io\image.py:13: UserWarning: Failed to load image Python extension: torch_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: <function _DenseLayer.forward at 0x000001F6781C65E0>. warnings.warn(f”Unable to retrieve source for @torch.jit._overload function: {func}.”) torch_jit_internal.py:751: UserWarning: Unable to retrieve source for @torch.jit._overload function: <function _DenseLayer.forward at 0x000001F6781C6820>. warnings.warn(f”Unable to retrieve source for @torch.jit._overload function: {func}.”) ‘’’
so I believe that maybe I have the wrong python installed or something like this? I’ve just unziped the files and executed the .exe as manager so in theory, it should work.
Exactly the same problem.
Could be, but im pretty sure its not. There is a setting to redner a second image, that could be it. Furthermore i am not having this issue, and i cant find any other evidence.
pretty sure you have a line break after your prompt, and it is generating an "empty" image because of the empty line... happens often when you copy your prompt from somewhere etc, or press "enter" at the end
I've had that happen a few times, it happens when you accidentally leave an empty line after the end of your prompts, like if you hit ENTER one too many times, or left a line after deleting a bunch of prompts.
4.5 stars.
saving the output folder would be very nice :D
Can't wait to try img2img here
I tried it on the patreon version and it's amazing! Hope it gets to itch for you guys as well soon :)
I've been using the NMKD GUI in the mean time
I have a Tesla m60, which works like an SLI card. Can we add functions like;
Use all GPUs or, Check with a marker with GPU (0), GPU (1), etc?
![]()
as you can see, GPU 1 is not using it's resources,
With that maybe I can reach full of 16 GB's of Vmemory
Hey, is there any way to remove the NSFW filter? (asking for a friend)
nsfw filter?
They want to generate...” content” with this. NSFW content.
I'm fairly certain its disabled by default, I didn't have to change or disable anything in the program.
"asking for a friend"
There is no NSFW filter here, though I'd assume the Stable DIffusion model (out of GRisk's control) probably isn't trained on anything more lewd than nudity.
Id like a direct download link and a version packaged as .zip please! also, cli mode is broken... but its an awsm tool!
could you tell me the paramenters for this beautiful woman? text, seed, etc..?
beautiful girl longshot, red hair, flower crown, hyper realistic, pale skin, 4k, extreme detail, detailed drawing, trending artstation, hd, fantasy, d&d, realistic lighting, by alphonse mucha, greg rutkowski, sharp focus, elegant
'steps': 60,
'vscale': 8.0,
'seed': 3314742242,
'resX': 512, 'resY': 704, '
thanks mate, you are a real hero!
Kind of funny stuff, version 0.3 made her smile! with the same prompt and seed, thanks for sharing btw!
thank you so much im so new to this changed her hair made it curly and blonde
erry girl longshot, blonde curly hair, flowers crown, hyper realistic, pale skin, 8k, extreme detail, detailed drawing, trending artstation, hd, fantasy, d&d, realistic studio lighting, by alphonse mucha, greg rutkowski, sharp focus, elegant with graffiti background model shoot, apocalypse took like 10 mins
do not recommend seed 3314742242 steps 50
Getting the following error: RuntimeError: CUDA out of memory. Tried to allocate 30.00 MiB (GPU 0; 2.00 GiB total capacity; 1.59 GiB already allocated; 17.26 MiB free; 1.62 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
My VGA card is a GeForce GT730, can i do anything with it?
Hi the only way to do this would be to try and generate the image at a smaller size. the vram recommendation is 4gb min and 6gb to generate images at the standard size of 512 by 512
This error has nothing to do with GPU, people with 12GB and 8GB of vram are also facing same issue. and reducing the size did not help
Did you manage to fix it? If so, please share your solution.
rtx 2060 super 704x512
Steps 60
V Scale 8.00
V
hope you can make that runs in Mac, awesome works, thanks
I also would really appreciate a Mac version. I tried running it in Parallels and it did not work.
hope the make on for Mac
This program requires Nvidia cards so it will never run on a Mac and AMD doesn't care about artificial intelligence (unfortunately for them) so cry yourself a river.
Of course you can't! You are trying to fool the program BUT it REQUIRES Nvidia hardware! How technologically ignorant can you be?
man i have a 1660. man pls just pls get the half precision thing done next update. this is not a demand. however it's a beg.