InvokeAI - The Stable Diffusion Toolkit
A downloadable tool for Windows
Requires a Nvidia Graphics Card (except GTX 1650).
Stable Diffusion - Write anything, the AI paints it!
No censorship, unlimited pictures, use them for anything you want, 6s per image on a RTX 3060.
- No internet connection needed.
It runs entirely on your PC.
Click here to learn more about stable diffusion.
InvokeAI - Powerful & Easy UI
- Modern & easy to use.
UI with the workflow in mind.
- So. Many. Features.
In/outpainting, img2img, model merging, face restoration, upscaling, ... There's barely anything InvokeAI cannot do.
- Try more art styles!
Easily get new finetuned models with the integrated model installer!
- Let your friends join!
You can easily give them access to generate images on your PC. Or use...
- Tablet mode!
Run InvokeAI on your PC, and connect your tablet/smartphone to it. Generate images from your bed!
- Free & open source.
Click here to learn more about InvokeAI.
About this standalone
Just download, unzip, double-click. No command line, no python installation needed.
- All the important models.
Comes pre-packaged with SD 2.1, SD 2.1 768px and SD 1.5! Also there is an integrated model downloader for the most famous models.
- Open Source.
You can find the source code in the download.
- Why this standalone?
This standalone is not affiliated with InvokeAI. I created it, because the normal InvokeAI installation can be tricky for new users. If you encounter problems with the standalone, join the official InvokeAI discord and ping @Sunija#6598.
♥ Share it with your friends! ♥
Need help with the installation?
Join the InvokeAI Discord and ping @Sunija#6598.
|Updated||8 days ago|
|Tags||ai, Generator, Graphical User Interface (GUI), image-generation, inpainting, invokeai, model, outpainting, stable-diffusion|
|Code license||MIT License|
|Average session||About a half-hour|
|Languages||German, English, Spanish; Latin America, French, Italian, Japanese, Dutch, Polish, Portuguese (Brazil), Russian, Ukrainian, Chinese (Simplified)|
|Multiplayer||Server-based networked multiplayer|
Log in with itch.io to leave a comment.
Hi, img 2 img doesn't work, always get an error up by the model, and it fails. Command prompt has this at the end each time.
"command = parameters_to_command(generation_parameters)
File "E:\Itch\Invoke_Ai\invokeai\invokeai_2_3_0_standalone\env\lib\site-packages\invokeai\backend\modules\parameters.py", line 45, in parameters_to_command
if "init_img" in params and len(params["init_img"]) > 0:
TypeError: object of type 'Image' has no len()"
Suffice to say I have put in an image into the prompt and also tested various settings, result is the same seemingly regardless of model used.
Of note, Unified canvas still works, so it can still generate based on another image, just something going wrong with img 2 img specificly.
Any chance the ControlNet or OpenPose add-ons?
Why I faced "An exception has occurred: 'cp950' codec can't decode byte 0xe6 in position 474: illegal multibyte sequence" problem ?
Is there any solutions?
Can you send me a screenshot of the error on Discord (Sunija#6598)? :)
I tried some images and they all came out blurred with a big exclamation mark.
what do I do?
turn off nsfw filter.
Are you ever going to add support for AMD gpus? It really sucks not being able to use this on my 6600XT. 😭
i have installed it but what should i do now:/ - can someone help me i am to dumb for that😂👍
This video should explain everything. :)
I read this line at the beginning of the threat: "Requires a Nvidia Graphics Card (except GTX 1650)."
what's this mean, does it mean that GTX 1650 is not gonna work lmao. I use 1650 lol:")))
please answer me thanks
It won't work. :(
The 16xx generation has a bug that makes it need more VRAM than other cards. The 1650 is the only model that doesn't have enough VRAM to do that. :(
Worst case scenario, use your cpu, you'll be able to render mostly anything, it will just take a long time...
Not to lead away from this program, but NMKD Stable Diffusion (here on Itch) has settings for low-memory cards as well as AMD cards.
I guess its not really a lead away if they literally cant use this haha.
So, today marks it one week since i downloaded this and I got to say, i only slept 12 hours in one week. I've been stuck to my computer ever since. After 3 days of using the cpu to render a picture in 55minutes, I went and I bought a 3060 without even looking at the money(lucky the card was 150$ off on sale). My only regret is that they didn't have a 3070 in stock. I just want things to load faster!! I'm considering of adding another gpu, maybe even an older one as long as it runs the latest cuda drivers.
One question: i would like to compile my own sampler, where could i find the source code for the existing samplers ?
Not to forget, a huge THANK YOU bro, for making this software real.
Cool that you like it! :)
I don't know where to find the samplers in the source code, but I bet the InvokeAI team can help you on their discord. :) Also, they are working on the big 3.0 update that will allow plugins/nodes, so adding a new sampler should be pretty easy by then. :)
And for AI the 3060 is usually better than the 3070. :) The 3060 has 12 GB of VRAM (compared to the 8 GB of the 3070), which allows for *much* bigger images than the 3070 would. :)
Do you happen to know if i add an extra card, will it help with the rendering time ? or it will just render a 2nd image on the side ?
Yo I found out about this project yesterday and used it a couple hours already, pretty good work, btw I already read a comment where you answered that the Loras support will be added on the next big release which was my only question so nice job overall, really looking forward to this and is it okay to try to get involved in this somehow? I'm a software engineering student and this really caught my attention, best wishes!! Hope to see more of this :)
You can join the InvokeAI discord, I bet they are always grateful for help! <3
Hey my windows defender has been detecting trojan wacatac b ml on your requeriments_checker, what is this?
Thanks for the report! :)
Can you send me a screenshot of the report on Discord (Sunija#6598)? My windows defender doesn't find anything there, so I cannot check myself. :(
And are you sure that it's the requirements checker? :3 There is a known issue that some of the AI models are falsely flagged as threats.
Also, in case somebody is interested, that's the whole script of the requirements checker / starter: https://github.com/GothaB/invokeai_starter/blob/main/invokeai_starter/MainWindow...
i sent you a friend request so i can show you the print
Hi! Will the DPM++ 2M Karras sampler be added?
Already in there! :)
Latest version gives me a "ImportError: DLL load failed while importing cv2" and won't start
Thanks for reaching out! :)
Can you send me a screenshot of the error on Discord (Sunija#6598), so we can fix it together? :)
Until then, you can start the AI via the "helper.bat" file next to the starter. :3
It's a great launcher but i do feel like it's mainly for beginners in ai art because as someone who is familiar with webui1111 there are some few things i dislike/like about it.
What i don't like about it:
1) No ability to read metadata on an image that wasn't generated inside the program (couldn't find the tab to do so anywhere)
2) Low quality upscalers , yes there are the options to 2x and 4x an image but the upscaler just isn't good and a good upscaler is a lot more important than the initial generated image. (Nearest, latent, 4xultra, animesharp.. all missing)
3) Loras, can't load loras
4) Ability to refresh embeddings once the program is already working
5) Sampler names are all shortened which i don't understand how anyone is supposed to know what is k_lms/k_huen/etc...
What i do like about it:
1) Generates images really fast (4s on a 512x512 image on a gtx 2070 super)
2) Unified canvas (amazing feature)
3) The UI in general is very comfy looking and inviting, it feels like having your own private Novel AI
4) Lots of tabs to help with image generation
5) You can load the model plus the chosen vae of that model without having to do so manually
In summary for beginners this is a 9/10 straight out great but for people who are accustomed to webui1111 high quality upscaling and lora usage this is a 6/10.
Can't wait to see more, and I'm sure one day it will be a 10/10! :)
Thanks for the detailed review! :)
I'll let the InvokeAI team know about it. LORAs will come with the next big upgrade, but e.g. I didn't think about the upscalers, so your comment definitely helped! :)
hey so i love what you did here it makes this an easy and accessible way to use AI art but i was wondering if there was a way to change my default model i like stable diffusion but i use anythingv4.5 more if you can please help me out i would appreciate it
Top right in the InvokeAi webui, there's an isometric cube icon for "Model Manager".
Click that, you'll see "+ Add New".
Right click in the relevant Windows folder path containing your model(s), and select "Copy Address".
Ctrl-V to paste into the "Checkpoint Folder" field and hit Enter.
You should now be able to add your model of choice, and all going well, it should then appear in the drop down, to the left of the Model Manager icon.
Hope that helps ;)
Soooooo, first thank you sincerely for both InvokeAI and Aiimages :)
Having some tech-diffs, to put it mildly - actually a bit of a pre melt-down panic level event is fast brewing! I have been using InvokeAI v4 for a few weeks, not updated to v5 yet, and tonight I opened up all 3 instances of Aiimages (using different models per instance to save on hassle of model switching) that had in the region of 60k+ generations between them. Intead of my vast wonderful galleries which were all there yesterday, I was met with 3 completely blank galleries and each one populated with the orig InvokeAi prompt - Content: cute owl, Style: watercolor painting.
PLEASE tell me that someone else has previously encountered this issue and there's a way to retrieve the galleries. The generated images are still there in the Output folders and both dream_log.md and dream_log.txt are still intact and contain all of my prompts for all 3 Aiimages instances I've been working in. Anyone got a fix for this? Otherwise I've just lost a vital chunk of half a years work.....
Kind dev, please let me know if this catastrophe can be fixed somehow. <3
Thanks for reaching out!
InvokeAI or Aiimages? :X First the good news: As long as the images are still there, nothing is lost on the InvokeAI site. Just to be sure: Make a backup before you continue. :3
You can set an output folder in the bottom left of the starter. Set this to the folder that InvokeAI should use (= the one where your outputs are).
If you really meant Aiimages, then we can look at that again. :) If you need more help just poke me on Discord (Sunija#6598) or write here. :)
Thanks for the reply!
Sorry, I appreciate why that might have been confusing, I was a little panic-stricken as I was typing last night!
Are you aware of the disappearing history issue in Aiimages (possibly caused by a crash or by opening Aiimages and InvokeAI at the same time), and have you or any of the users on Itch or Discord found a method that would make it possible to restore my history in each of the Aiimages instances?
I think I may also have accidentally opened both InvokeAi and then Aiimages at the same time last night, which I've been avoiding, as from past experience, I know there's a shared file hiding somewhere outside the Aiimages folders, and have seen prompts and settings being populated between each instance of Aiimages when switching the instance of Aiimages I'm working in. I suspect that may be the issue, as upon opening Aiimages presently, the initial settings for InvokeAi appear, with the "cute owl" prompt loaded, and all the images generated are no longer there in history.
Thank you for reading my lengthy reply(s)! Hope my situation is more clear
Post Coffee Edit: Separated feedback and observation for both programs in to a separate post below :)
Currently I'm using both of your standalone offline tools. I have been generating material in Aiimages over the last half a year, and this month have started using InvokeAi mainly for the Unified Canvas feature. Taking images made in Aiimages, and importing them to InvokeAI to develop them further with the canvas toolset.
I have three instances/installs of Aiimages that I use for separate content/workflows using different models, and also because after 20k+ images are in history, in my experience it gets a little slow and frequently crashy when scrolling through the vast galleries.
With InvokeAi, I find its design more of a hassle for accessing earlier generations because it only loads 50 (or is it 100?) images in to cache, and then I have to constantly scroll down and click to load another 50/100 images. I generate LOTS of images per prompt so accessing the images becomes cumbersome in InvokeAi. Very klunky and limiting in contrast with Aiimages, in that respect. Pros and Cons yo!
For my workflow, being able to load up one of the Aiimages instances and quickly scrolling the history to get the precise prompt for an image to work on inside InvokeAi's unified canvas has been invaluable. So while I still have the images, and still have the dream_md / .txt prompt histories for each of the three Aiimages instances intact in their respective folders, I can no longer easily scroll the history to access the image with corresponding prompts.
Aiimages handles the Style/Content fields a little differently to InvokeAi with its Prompt/Negative fields. For instance, if I copy a Style/Content prompt from Aiimages into InvokeAI's prompt field, all is well until I reload the prompt at a later stage, as it then takes the original prompting character/token sequence and populates the Negative prompt field with any [negatives] from the original Aiimages prompt. It appears to modify the generated output aesthetic quite dramatically at times when this happens, so I usually just copy the prompt from Aiimages back in to the InvokeAi prompt field each time I want to return to a theme (say for going past the 1024 x 1024 limit in Aiimages) to keep the precise prompt.
If I'm not mistaken, this also means the metadata when taking an image from one application to the other changes the numerical(/character?) string that the prompts are converted into, so I very much hope to retain the history of images generated in my Aiimages instances over the last 6 months inside Aiimages for easy access of/reference to the original prompts.
Thanks so much for releasing these tools for the beginners, the Unity devs in the case of Aiimages, and for the folk who just like to run their machines offline to avoid the overlords snooping and auto-update tampering with their delicate work environments! :D x
Thanks for the detailed review! :)
First the bad news: There is no easy way to restore the aiimages history file. :X
The good news: All those images still have their prompts/settings in the meta info. And InvokeAI can read those meta infos.
So I guess the best way atm to reproduce your workflow would be:
1) Open your aiimages output folder in your Windows Explorer, set the images to maximum size.
2) If you want to use the prompt of an image, copy it into the invokeai output folder.
3) In InvokeAI, you can click on the image and use the "use prompt" button on top.
Let me know if that workflow works for you. :X
In theory it could be possible to scan a folder, read the metadata and rewrite it into an aiimages history file. But that would be quite a bit of work for a deprecated tool. :X
Hey thanks so much for your reply. :)
As a temp workaround that could well be the solution, good thinking and thanks for the suggestion!
Sad/bad news indeed on all fronts, especially re: Aiimages' deprecated status D:
I absolutely love the design of Aiimages, and as a Unity dev I'm still planning to integrate it in to some of my projects. In the Unity space it's still IMHO the best SD tool available, by far, and I've tried them all (except Keijiro Takahashi's recent addition for the Mac folk). Props are due for the work you've put into it!
My waifu is fairly experienced with Python. If you have a sense of what would be involved to scan and rewrite the history file, would you be ok with sharing your pseudo-code thoughts on what steps would be required in Python to achieve the history restoration?
If we have any luck achieving it, we'd of course share the scripts with the community in case anyone else encounters the issue.
Many thanks again Sunija! <3 x
Hi! When I try to load the image I get this message:
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 MiB (GPU 0; 6.00 GiB total capacity; 4.34 GiB already allocated; 13.56 MiB free; 4.41 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
>> Could not generate image.
>> Usage stats:
>> 0 image(s) generated in 3.99s
>> Max VRAM used for this generation: 4.66G. Current VRAM utilization: 4.66G
>> Max VRAM used since script start: 4.66G
The image doesn't load and it stays grey :< Is there anything I can do? I have GeForce GTX 1660 Ti
Sounds like you're low on VRAM. :3 There are several things that you can try:
1) Close InvokeAI and check how much VRAM is already used up by the system in your task manager. Try to close stuff that you don't need and check if it lowers the VRAM usage.
2) Disable the NSFW checker, as it also needs some VRAM. You have to restart InvokeAI so the setting is applied.
3) Disable the in-progress preview, because it also needs VRAM. You can disable it in the tool on the cogwheel in the top right -> Display In-Progress Images -> None
4) There is a --free_gpu_mem option that will save VRAM. It will slow the generation a bit, though. To activate it, you have to open invokeai_2_3_0_standalone\invokeai\invokeai.init in a text editor and add the line after one of the upper rows. But I'd first check if the rest works. :3
Just a side note, after reading about Windows antivirus below. Bitdefender saw two of the Python files as Ransomeware. I added then them as OK and the install continued as normal.
HI. Unfortunately, when I launch Invoke AI I get an error, specifically PermissioneError WinError 10013 Attempt to access the socket in ways not allowed by the respective access permissions. I don't know what to do unfortunately
Can you try to run it as an administrator? :)
(Rightclick -> Run as administrator)
If that doesn't help, poke me on Discord (Sunija#6598) so we can look for a solution. :)
Thanks for availability, my discord is ElPolloLoco#0042, I tried to run it as administrator, disable windows defender and windows firewall but it still doesn't work
Hey! Great job, it's super convenient and easy!
Can we expect LoRAs support soon? If yes, then roughly when? Around April? Thank you for your work!
I keep getting a Down Load Issue at 954Mb, I've tried 3 times. It could be a local issue.
If you download via the itch.io/app then you should be able to resume the download. :)
You were 100% correct, downloaded and installed beautifully!!
I'm having a problem with the easynegative embedding and as the only thing I found about this is another comment, i'm trying my luck too:
>> Notice: embeddings/easynegative.pt was trained on a model with an incompatible token dimension: 768 vs 8
The embedding does not show in the list and does not seem to do anything when I try to use it anyway. (No ">> Preparing tokens for textual inversion")
Guys, if your trying to use an AMD gpu it will not work on windows (for now). But if you want to use stable diffusion i cant recommend Google colab (Stable diffusion) enough.
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???
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!
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.
- 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?
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)?