The 3060 should be better. :) On my desktop 3060 a 512x512 pixel image takes 5 seconds. The VRAM size does not affect speed, but limits to to smaller images (though with 6 GB you should be fine :) ).
When I click on start it shows the console but it closes immediately, and it keep saying "Loading" in the button but nothing happens at all, it just takes that long or it's a problem?
Thanks for reaching out! Can you try double clicking the "helper" file in the same folder? :3 If you installed via itch, you can open the folder via the cogwheel button next to the launch button -> manage -> open folder. :) If that doesn't help much, poke me on Discord (Sunija#6598).
Hi! How does share access actually work? I have tried using the "Your Wifi" link on my phone yet all I get is "This site cant be reached , took to long to respond." I am also running it on my pc at the same time, what am I missing?
Thanks for reaching out! :) Did you restart it after activating the "share access" setting? :3
Other common problems are: - The phone is not in the wifi. - The windows firewall is blocking it. When you start InvokeAI with shared access for the first time, windows should ask you if you allow it to go through the firewall.
After uninstalling with the Itch launcher, CTRL+SHIFT+N on my browser now longer opens a private window and instead gives the error: "Missing Shortcut - Windows is searching for invokeai_starter.exe."
It will also run on the amd ryzen 5, but it will be really slow without a nvidia graphics card (5 min per image, instead of some seconds on the nvidia card).
Hi! I've got 2 problems which have surfaced when trying to download this.
1, Windows Defender and other various programs are detecting the Backdoor Program "Bladabind!ml" in one of your bin files, this can lead to serious damage if not dealt with and I had to uninstall and wipe those files. 2, If the Backdoor program is not an issue, I own a AMD Radeon RX 6800 and was wondering if this program would be compatible, as I've tried to run it but no use. Thanks!
1) The threat alert should not be an issue. ~3 days ago Windows Defender started making various AI files as threats that have been used for months. I posted a longer explanation some comments below. :)
2) Sadly, AMD cards are not supported yet. :( Generation will run on your CPU then, so it will be really slow (5-10 minutes per image).
Hi, is this really safe to use right now? Windows Defender keeps telling me there are some trojans and a backdoor in the files. I can just ignore it but is there something else we can do about it right now?
Your standalone version is amazing but I had to stop using it because idk what to do about the whole virus situation.
TL;DR: Yeah, from all that we know, it should be fine to use.
More info: - 2 days ago windows defender started marking various ai models as threats - no other antivirus program sees a threat there (checked via virustotal.com) - there was no change to the models. Those models have been uploaded and used for months and come from their official sources on huggingface - this is not only happening for the standalone, but also normal InvokeAI, auto1111 and just random vaes
Our best guess is: Windows Defender has to use a quick and imprecise method to check all the files on your system without slowing it down. So it most likely mistakes those official models for shadier ones.
Sooo, we just have to wait for Microsoft to fix that. :/ But I'll also check how to add exceptions to the Windows Defender, so it doesn't eat your models and breaks InvokeAI. Or breaks my InvokeAi. :/ I couldn't test the newest patch (which doesn't change the models) because Defender keeps deleting my models after I copied them somewhere else.
● won't be able to use InvokeAI's inbuilt converter (via command line) - that is, you won't be able to update your old models to the newer "diffuser" format.
● won't be able to run the 'batch render' to automate a quantity of renders, so you'll have to sit and enter data every minute for hours on end.
Quick & easy install method with problems down the track ......
Actually, this is a full InvokeAI install, so you can still use the command line. :3
The "helper.bat" basically calls the InvokeAI installation via the command line. So you can make a copy of that one and adapt it to your needs. :)
Edit: You can update models to diffuser from within the UI, when you load it normally -> Edit -> Convert to Diffuser. The button is a bit hidden, sadly. :/
When I install and run it from the itch.io application, the application appears to be running but does not open. what is the reason? (i will try directly download through website this time ) . and another problem; during the installation windows defender caught 2 trojans and a backdoor called win32/Bladabind!ml. does this cause a problem about the security of my device? (interestingly I did not get any virus detection when I scanned the same files in virustotal).
You can try to run it directly from the folder (cogwheel next to launch button -> manage -> open folder) and then double click invokeai_starter.exe. After that, you should have a desktop shortcut to run it.
(I don't really know why itch takes so weirdly long to launch it. :/ Maybe it is checking if all files are there, which could take a while for 21 GB.)
2. It is working for me. :X Are you sure you use the right trigger word? The trigger word might be different from the file name or what somebody wrote in civitai. InvokeAi will write a list of trigger words in the console after it loaded a model, e.g. ">> Textual inversion triggers: <low-poly-hd-logos-icons>, etc". There you find the actual trigger words.
1. There is a limit of about 75 tokens, so if the negative textural inversions work, that's usually a good way to save on words. :)
Getting backdoor:win32/Bladabind!ml from windows defender. Multiple users getting this after the last backdoor was picked up. seems to have been changed
Windows defender started detecting various non-threatening ai files as threats.
Some info: - If you check the files via virustotal (which checks it with 58 antivirus programs) it doesn't find a threat. - It is not limited to the standalone or InvokeAI, but apparently also e.g. deleting vae's from the auto1111 ui: https://www.reddit.com/r/StableDiffusion/comments/11vo5ke/warning_windows_defend... - This can block the installation, because windows might delete files right when they are downloaded. This doesn't happen on all systems though. - There was no patch that changed the files. Last patch was 5 days ago (as you can see in itch) and reports of the problem started yesterday. So apparently windows just got an update.
TL;DR: We'll have to wait until Microsoft fixes the defender. :(
When you add a diffuser model, it will ask you directly to put a vae. :) If you add a safetensor/checkpoint, you first have to add it, then click on the "edit" button next to the model in the model list. Then you can add your vae there. :)
Hi, thanks for reaching out! :) You have to rename your "Invoke AI" folder to "InvokeAI". The space in between kills the program. :3 Sorry for the inconvenience!
Hey! I tried installing it on itch.io, but it clearly seems within the files installing was as trojan called Blababindi... I did some research, it's pretty dangerous, and I wonder how it could have gotten inside these files... Fortunately Windows Defender quarantined it, but still... Would you have an explanation (no it's not a bug there was indeed Blababindi!ml inside of it)
To be exact here's the address from the itch folder in %appdata% : "itch\apps\invokeai\invokeai_2_3_0_standalone\invokeai\models\hub\models--CIDAS--clipseg-rd64-refined\snapshots\583b388deb98a04feb3e1f816dcdb8f3062ee205\pytorch_model.bin"
Hmmh ... Maybe.... I don't doubt your honesty, since it's all open source, but scary stuff though, I think I'll install it on a VM. Just to be sure. And not from itch.io
I can't run the program for some reason, it keeps spitting out this error: OSError: [Errno 22] Invalid argument: 'D:\\itch\\invokeai\\invokeai_2_3_0_standalone\\invokeai\\models\\hub\\models--runwayml--stable-diffusion-v1-5\\snapshots\\ded79e214aa69e42c24d3f5ac14b76d568679cc2\\safety_checker\\pytorch_model.bin'
i don't know what to do. If i ignore the error and procede to launch the program, it'll start and open the web ui but i cannot do anything on it or it'll crash. i've reinstalled it 5 times at this point.
i have a 10700K and an RTX3050.
edit:
If i reinstall the models as suggested it gives this error instead:
CacheNotFound: Cache directory not found: D:\itch\invokeai\invokeai_2_3_0_standalone\invokeai\models\diffusers. Please use `cache_dir` argument or set `HUGGINGFACE_HUB_CACHE` environment variable.
Sadly not. :( The 16xx have a bug that can only be circumvented with additional vram usage. The 1650 is the only model that doesn't have enough vram for this fix.
I wonder in cases of VRAM limits, doesn't windows as well as many videocards allow for shared memory use? Seen it done in other areas but not with things like stable diffusion Basically if shared memory can be used, then core functions can still work even if it will be really slow.
For example, allowing things as far as a worst of all worlds memory setup (cursed image of a GTX970), as soon as a task starts to need more than 4GB it can gradually stop using that last 512MB pool to avoid context switching there, and start using a bunch of system RAM.
If you use the itch app, it will update automatically. If you don't use it, download the new version and copy your output folder from the old version to the new one. After you copied your output folder (or made a backup of it), you can delete the old version.
I tried to launch the application but it did not load or run the application at all. I am using rtx 2060 graphic card and AMD CPU ryzen 5600G, is there anything that I am missing?
Thanks for reaching out! :) Are you probably using a vpn?
(As background: To display your IP, the starter has to contact api.ipify.org. This doesn't work if you're behind a vpn. It definitely shouldn't crash because of that and that will be fixed with the next patch. You can test if it's crashing because of that by just running the helper.bat file in the invokeai folder. :3)
I think it will be possible. On linux it works on AMD and there is a plugin for the auto1111 ui which supports amd gpus, apparently. The latter one is harder to install, though. :/
But I really cannot tell how long it'll take until we get an AMD gpu version. I thought it would have been done much sooner, but there's now this plugin for another UI... so - as far as I know - it might be just a few weeks away or several months. :/
On my 10 year old FX 8350 cpu, it took 8 minutes to generate a 512x512 image. On yours maybe 3-4 minutes, but I don't know.
So it won't be super fun, but if you take a good model/prompt from civitai and leave it running while doing something else, it miiiight be okayish.
Ok, great. I figured it wasn't because I'm only getting 7-8it/s on my brand new machine with a 4090. I noticed others reporting 13-14 on reddit with the same card. Do you have any suggestions on optimization settings?
Thanks for the convenient InvokeAI version, is it possible in future versions to reduce the number of installed models from 4 to 1 or even 0? The weight of four models is too big, I think it would be convenient to provide a light archive
Hm... we could go from 21.4 GB to 14.4 if we only leave 1.5 (no 1.5 inpainting, no 2.1 models). I tried to pack all the models that would require a huggingface api key to download.
Btw. you should be able to delete those models to save up space. :3
← Return to tool
Comments
Log in with itch.io to leave a comment.
Do you have any plans to publish this software onto cloud? Similar to DALL-E and Midjourney?
When i put to download Invoke in my ccleaner browser it tells me it is prohibited, is it because of cleaner?, it acpted well the checker.
will my rtx 3060 laptop gpu with 6gb of vram be enough?? :(, and how long will it takes to load an images with 6gb of vram
or will it better if i use Gtx 1660 TI 6gb in my desktop instead?
The 3060 should be better. :) On my desktop 3060 a 512x512 pixel image takes 5 seconds.
The VRAM size does not affect speed, but limits to to smaller images (though with 6 GB you should be fine :) ).
We need Mac version too. :/ Interface looks amazing
You can install it on Mac with the automated installer from here: https://github.com/invoke-ai/InvokeAI#getting-started-with-invokeai
But I hope we'll also get a standalone for Mac. :X
hi, I have am AMD video card, 6800 XT, will it work? Also, how can I load models? Thanks!
Use "invokeai requirements checker" for checking yor GPU and here is tutorial on models
will there be amd gpu support coming?
I think yes, but I don't know when. :/
that would be great^^ my cpu can handle the rendering (AMD power) but using my gpu would be cool too^^
When I click on start it shows the console but it closes immediately, and it keep saying "Loading" in the button but nothing happens at all, it just takes that long or it's a problem?
Thanks for reaching out!
Can you try double clicking the "helper" file in the same folder? :3 If you installed via itch, you can open the folder via the cogwheel button next to the launch button -> manage -> open folder. :) If that doesn't help much, poke me on Discord (Sunija#6598).
Hi! How does share access actually work? I have tried using the "Your Wifi" link on my phone yet all I get is "This site cant be reached , took to long to respond." I am also running it on my pc at the same time, what am I missing?
Thanks for reaching out! :)
Did you restart it after activating the "share access" setting? :3
Other common problems are:
- The phone is not in the wifi.
- The windows firewall is blocking it. When you start InvokeAI with shared access for the first time, windows should ask you if you allow it to go through the firewall.
After uninstalling with the Itch launcher, CTRL+SHIFT+N on my browser now longer opens a private window and instead gives the error: "Missing Shortcut - Windows is searching for invokeai_starter.exe."
How can I get rid of this shortcut override?
The ctrl+shift+n comes from the desktop shortcut, so if you delete it, it disappears. :3
If you reinstall, the ctrl+shift+n thing does not appear anymore. :)
Will this run on a ryzen 7 and 3050 ti? And is it only gonna work on pc or a yes for latops?
It will work on a 3050ti and it doesn't matter if laptop or pc. :3
hi i just wanted to ask will this run on an amd ryzen 5 5600g???
no
would it run on a i7 7700k, 16gb ram and a gtx1080?
It should. :3
It will also run on the amd ryzen 5, but it will be really slow without a nvidia graphics card (5 min per image, instead of some seconds on the nvidia card).
Hi! I've got 2 problems which have surfaced when trying to download this.
1, Windows Defender and other various programs are detecting the Backdoor Program "Bladabind!ml" in one of your bin files, this can lead to serious damage if not dealt with and I had to uninstall and wipe those files.
2, If the Backdoor program is not an issue, I own a AMD Radeon RX 6800 and was wondering if this program would be compatible, as I've tried to run it but no use. Thanks!
Hi!
1) The threat alert should not be an issue. ~3 days ago Windows Defender started making various AI files as threats that have been used for months. I posted a longer explanation some comments below. :)
2) Sadly, AMD cards are not supported yet. :( Generation will run on your CPU then, so it will be really slow (5-10 minutes per image).
my windows defender found trojan virus inside the folders
Hi, is this really safe to use right now? Windows Defender keeps telling me there are some trojans and a backdoor in the files. I can just ignore it but is there something else we can do about it right now?
Your standalone version is amazing but I had to stop using it because idk what to do about the whole virus situation.
TL;DR: Yeah, from all that we know, it should be fine to use.
More info:
- 2 days ago windows defender started marking various ai models as threats
- no other antivirus program sees a threat there (checked via virustotal.com)
- there was no change to the models. Those models have been uploaded and used for months and come from their official sources on huggingface
- this is not only happening for the standalone, but also normal InvokeAI, auto1111 and just random vaes
Our best guess is: Windows Defender has to use a quick and imprecise method to check all the files on your system without slowing it down. So it most likely mistakes those official models for shadier ones.
Sooo, we just have to wait for Microsoft to fix that. :/ But I'll also check how to add exceptions to the Windows Defender, so it doesn't eat your models and breaks InvokeAI. Or breaks my InvokeAi. :/ I couldn't test the newest patch (which doesn't change the models) because Defender keeps deleting my models after I copied them somewhere else.
*** WITHOUT COMMAND LINE Access *** - you;
● won't be able to use InvokeAI's inbuilt converter (via command line) - that is, you won't be able to update your old models to the newer "diffuser" format.
● won't be able to run the 'batch render' to automate a quantity of renders, so you'll have to sit and enter data every minute for hours on end.
Quick & easy install method with problems down the track ......
Actually, this is a full InvokeAI install, so you can still use the command line. :3
The "helper.bat" basically calls the InvokeAI installation via the command line. So you can make a copy of that one and adapt it to your needs. :)
Edit: You can update models to diffuser from within the UI, when you load it normally -> Edit -> Convert to Diffuser. The button is a bit hidden, sadly. :/
Is it possible to add new samplers on my own, for example, dpm++ 2m karras and others, or do I need to wait for the corresponding updates?
Many thanks!
dpm++ 2m karras = k_dpmpp_2
So, it is already implemented. :3 I don't think you can add others, though.
So we can't add DPM++ SDE Karras?
When I install and run it from the itch.io application, the application appears to be running but does not open. what is the reason? (i will try directly download through website this time ) . and another problem; during the installation windows defender caught 2 trojans and a backdoor called win32/Bladabind!ml. does this cause a problem about the security of my device? (interestingly I did not get any virus detection when I scanned the same files in virustotal).
You can try to run it directly from the folder (cogwheel next to launch button -> manage -> open folder) and then double click invokeai_starter.exe.
After that, you should have a desktop shortcut to run it.
(I don't really know why itch takes so weirdly long to launch it. :/ Maybe it is checking if all files are there, which could take a while for 21 GB.)
Okay, but that's the problem. starter.exe won't open. I've tried a lot of things, but for some reason it didn't work.
I have problem, when I click to launch it shows running, then nothing happens, again Launch button.
Thanks for reaching out! :)
Can you send me a screenshot of your error and your starter on Discord (Sunija#6598)?
1. My prompts getting truncated cause of length, any fix to that expect shorter prompts?
2. Can't put negative TEXTUAL INVERSION prompt, ai just ignores it, positive TE prompts wokrs as expected
2. It is working for me. :X Are you sure you use the right trigger word? The trigger word might be different from the file name or what somebody wrote in civitai.
InvokeAi will write a list of trigger words in the console after it loaded a model, e.g. ">> Textual inversion triggers: <low-poly-hd-logos-icons>, etc". There you find the actual trigger words.
1. There is a limit of about 75 tokens, so if the negative textural inversions work, that's usually a good way to save on words. :)
Ok,new issue, some of TI throw to me that warning and not diplays in list:
Notice: embeddings/easynegative.pt was trained on a model with an incompatible token dimension: 768 vs 8.
Getting backdoor:win32/Bladabind!ml from windows defender. Multiple users getting this after the last backdoor was picked up. seems to have been changed
Windows defender started detecting various non-threatening ai files as threats.
Some info:
- If you check the files via virustotal (which checks it with 58 antivirus programs) it doesn't find a threat.
- It is not limited to the standalone or InvokeAI, but apparently also e.g. deleting vae's from the auto1111 ui: https://www.reddit.com/r/StableDiffusion/comments/11vo5ke/warning_windows_defend...
- This can block the installation, because windows might delete files right when they are downloaded. This doesn't happen on all systems though.
- There was no patch that changed the files. Last patch was 5 days ago (as you can see in itch) and reports of the problem started yesterday. So apparently windows just got an update.
TL;DR: We'll have to wait until Microsoft fixes the defender. :(
How I can put in custom vae?
When you add a diffuser model, it will ask you directly to put a vae. :)
If you add a safetensor/checkpoint, you first have to add it, then click on the "edit" button next to the model in the model list. Then you can add your vae there. :)
Thank you, great work. I really satisfied with AI. Can't wait for the lora though) wish you best of luck. (Bad English alert)
Hi, I cant run the program. It starts to run then I get :
model stable-diffusion-1.5 could not be loaded: [Errno 22] Invalid argument
then
** trying to reload previous model
>> Loading diffusers model from runwayml/stable-diffusion-v1-5
| Using faster float16 precision
| Loading diffusers VAE from stabilityai/sd-vae-ft-mse
Fetching 15 files: 100%|███████████████████████████████████████████████████████████████████████| 15/15 [00:00<?, ?it/s]
| Calculating sha256 hash of model files
** An error occurred while attempting to initialize the model: "[Errno 22] Invalid argument:
** This can be caused by a missing or corrupted models file, and can sometimes be fixed by (re)installing the models.
Do you want to run invokeai-configure script to select and/or reinstall models? [y/N]:
I have tried the reinstall and I get the same error.
Thank you for reaching out! :)
Can you send me a screenshot of your error and your starter on Discord (Sunija#6598)?
Hi, I can't run the program. :(
This is what shows the CMD when I click Start.
"G:\Programas\Invoke" no se reconoce como un comando interno o externo,
programa o archivo por lotes ejecutable.
G:\Programas\Invoke AI\invokeai\invokeai_2_3_0_standalone/env/python.exe: can't open file 'G:\\Programas\\Invoke': [Errno 2] No such file or directory
Hi, thanks for reaching out! :)
You have to rename your "Invoke AI" folder to "InvokeAI". The space in between kills the program. :3 Sorry for the inconvenience!
Thanks, solved
is it normal when winows defenders find trojans and other threats when installing this?
Yes, this is a false warning and can be dismissed. Registering the program (so this false warning disappears) would sadly cost a lot of money. :/
The starter is open source, so if there was a virus included, people could easily find out. Source code: https://github.com/GothaB/invokeai_starter
Hey! I tried installing it on itch.io, but it clearly seems within the files installing was as trojan called Blababindi... I did some research, it's pretty dangerous, and I wonder how it could have gotten inside these files... Fortunately Windows Defender quarantined it, but still... Would you have an explanation (no it's not a bug there was indeed Blababindi!ml inside of it)
To be exact here's the address from the itch folder in %appdata% : "itch\apps\invokeai\invokeai_2_3_0_standalone\invokeai\models\hub\models--CIDAS--clipseg-rd64-refined\snapshots\583b388deb98a04feb3e1f816dcdb8f3062ee205\pytorch_model.bin"
And the Defender patch note : Backdoor:Win32/Bladabindi!ml threat description - Microsoft Security Intelligence
Most likely a mistake of Windows Defender. If you upload the file to virustotal, no other virus scanner detects it as a virus: https://www.virustotal.com/gui/file/dd9308225b8314bb7236f207e6ea72b22db5d90dba03...
I'll check how to contact Microsoft to figure that out.
good luck, when i made something on visual studio i contacted Microsoft and they got it rid in the next security intelligence update
Hmmh ... Maybe.... I don't doubt your honesty, since it's all open source, but scary stuff though, I think I'll install it on a VM. Just to be sure. And not from itch.io
This is being discussed on Github too. Different detection in a different release of pytorch_model.bin, but same problem.
hi!
I can't run the program for some reason, it keeps spitting out this error: OSError: [Errno 22] Invalid argument: 'D:\\itch\\invokeai\\invokeai_2_3_0_standalone\\invokeai\\models\\hub\\models--runwayml--stable-diffusion-v1-5\\snapshots\\ded79e214aa69e42c24d3f5ac14b76d568679cc2\\safety_checker\\pytorch_model.bin'
i don't know what to do. If i ignore the error and procede to launch the program, it'll start and open the web ui but i cannot do anything on it or it'll crash. i've reinstalled it 5 times at this point.
i have a 10700K and an RTX3050.
edit:
If i reinstall the models as suggested it gives this error instead:
CacheNotFound: Cache directory not found: D:\itch\invokeai\invokeai_2_3_0_standalone\invokeai\models\diffusers. Please use `cache_dir` argument or set `HUGGINGFACE_HUB_CACHE` environment variable.
please help me before i lose my mind.
thanks in advance.
Thanks for reaching out! :)
Can you send me a screenshot of the full error on discord (Sunija#6598)? :)
Sure thing, I’ll contact you tomorrow.
Btw i think what you do is amazing.
Not sure if it helps diagnosing the issue, but I ran across this because I'm experiencing the same, also installed on `D:` on windows. Using RTX 3080
Your virus scanner is probably removing pytorch_model.bin as soon as it's downloaded. It may be a false positive. See this thread.
ugh, you're right. Windows didn't notify me, but it removed it. Adding an exclusion for InvokeAI worked for me. Thanks!
Is it possible on 1650
Sadly not. :( The 16xx have a bug that can only be circumvented with additional vram usage. The 1650 is the only model that doesn't have enough vram for this fix.
Is it not possible as well on GTX 1650 Super?
It also won't work on the GTX 1650 Super, because that one also only has 4 GB of VRAM (and it would need at least 6). :(
Does it works on RTX 2060?
I wonder in cases of VRAM limits, doesn't windows as well as many videocards allow for shared memory use? Seen it done in other areas but not with things like stable diffusion
Basically if shared memory can be used, then core functions can still work even if it will be really slow.
For example, allowing things as far as a worst of all worlds memory setup (cursed image of a GTX970), as soon as a task starts to need more than 4GB it can gradually stop using that last 512MB pool to avoid context switching there, and start using a bunch of system RAM.
i love the ui... thankyou
how do I fix the error: ModuleNotFoundError: No module named 'ldm'
Thanks for reaching out! :)
Can you send me a screenshot of the whole error on Discord (Sunija#6598)? :)
Will you add the opportunity to use lora models? And if yes, do you already know when this could be implemented?
Loras will be added with the next InvokeAI update, but sadly I don't know when this will happen. :(
Thats good enough news for me :) Thank you for this wonderful software
there goes me with my gtx 1650
are u ever gonna make a verison for amd or is it not possible
Will this be upgraded to InvokeAI-2.3.2?
The newest version is already on 2.3.2. :3
If you use the itch app, it will update automatically. If you don't use it, download the new version and copy your output folder from the old version to the new one. After you copied your output folder (or made a backup of it), you can delete the old version.
Thank you.
Can we use loRa models?
Sadly, not yet. :3
Thanks for reaching out! :)
Are you probably using a vpn?
(As background: To display your IP, the starter has to contact api.ipify.org. This doesn't work if you're behind a vpn. It definitely shouldn't crash because of that and that will be fixed with the next patch. You can test if it's crashing because of that by just running the helper.bat file in the invokeai folder. :3)
I got the following error
Rename the folder "F:\New folder" to "F:\New_folder". :)
When can we expect it to run on amd gpu, if ever :( ?
Edit: Also will the performance be bearable on r7 5800h ?
I think it will be possible. On linux it works on AMD and there is a plugin for the auto1111 ui which supports amd gpus, apparently. The latter one is harder to install, though. :/
But I really cannot tell how long it'll take until we get an AMD gpu version. I thought it would have been done much sooner, but there's now this plugin for another UI... so - as far as I know - it might be just a few weeks away or several months. :/
On my 10 year old FX 8350 cpu, it took 8 minutes to generate a 512x512 image. On yours maybe 3-4 minutes, but I don't know.
So it won't be super fun, but if you take a good model/prompt from civitai and leave it running while doing something else, it miiiight be okayish.
You know what ? Maybe I just use linux then! Will it work if I have both OS on my system, like in dual boot ? Or will I have to uninstall windows ?
You won't have to uninstall windows. :3
Though, this standalone only works with windows. :( So for Linux you have to follow the automatic installer guide: https://invoke-ai.github.io/InvokeAI/installation/010_INSTALL_AUTOMATED/#walk-th... and the rocm (= "AMD cuda") guide https://invoke-ai.github.io/InvokeAI/installation/030_INSTALL_CUDA_AND_ROCM/#wal...
It's not crazy hard, but it's obviously a bit trickier. :)
hell yeah, unlimited cat girls!
How do we install xformers for this version of invoke?
It comes pre-installed.
Ok, great. I figured it wasn't because I'm only getting 7-8it/s on my brand new machine with a 4090. I noticed others reporting 13-14 on reddit with the same card. Do you have any suggestions on optimization settings?
safetensors?
They work. If that's the question.
Thanks for the convenient InvokeAI version, is it possible in future versions to reduce the number of installed models from 4 to 1 or even 0? The weight of four models is too big, I think it would be convenient to provide a light archive
i feel you
Hm... we could go from 21.4 GB to 14.4 if we only leave 1.5 (no 1.5 inpainting, no 2.1 models). I tried to pack all the models that would require a huggingface api key to download.
Btw. you should be able to delete those models to save up space. :3
Is there a way to launch this app with a 2gb video memory?
It will not use your graphics card but your CPU then. That will be a lot slower than running it on the graphics card (a couple of minutes per image).
The AI already takes 2.5 GB in your video memory, and the image would take some more then. So 2 GB is not enough to run it on the graphics card.