You can increase the maximum step count by... 1) Options -> open settings folder 2) Close aiimages. 3) Open settings.json with Editor. Edit the line ""iMaxStepCount": 150,". Save it. 4) Open aiiamges.
Sorry that there's not easier way yet. :( Will come in the future. <3
Thank you for your reply. Can I also ask if there is a way to choose a custom output folder different from the default one? I wanted to use the software in an M2 and output it on a external HD
I've been using this thing non-stop like 5+ hours a day. It just is so much fun.
I've spent the past day or so reading up on the embeddings/textual inversion and google colab/Dreambooth options as well to train the stable diffusion model more, but it seems every option becomes difficult to various reasons - and the other options require you to pay to run a server farm to train the model for you.
Is there any chance you are working on or can put together a UI so we can train and create our own model.ckpt file directly in the software? I've looked at other options but require installing Ubuntu in Windows, or having to go through all that difficult Python scripts when we just want to be able to drag and drop images, click go and come back to a sweet new file.
Please, if you could just include a simple UI it would mean the world to myself and so many others. Your option to generate unlimited images makes it far better than the other stable diffusion UIs.
I think you cannot train an own model.ckpt on consumer hardware. :( And it needs about 40k images with labels. But there are two things I might add.
A) Embeddings. You can find those in the documentation for invokeai. Basically: Give the ai "5 images + 1 tag + 1h training on a RTX 3090ti = ai learned new word". I could create a UI for that already.
B) Dreambooth. I don't understand it yet, but it's kinda "embeddings but better", if I understood it correctly. It is not implemented in the backend yet, but it will be implemented. I'm kinda hyped for that, because it allows to create the same person in different poses. <3
Pretty impressed by what it can spit down with a nice prompt and some fine tuning. It can even manage some pretty face and hands instead of the distorted monstruosity that i saw in some other browser based generator.
It seem to get stuck on an infinite loading while i try to upscale or repeat an already upscaled / repeated image, but still for a generous price tag of free is one of the best tool that i've ever seen
Thank you. All other functions work normally. Only in the inpaint, if you want to modify the image partially, it will be stuck all the time. Why? （3060ti）
recently ive seen some rlly cool ai animation videos that show a bunch of ai generated images together like an animated video using stable diffusion and I was just curious if its possible to do with this?
would greatly appreciate some kind of integration or pipeline for Textual Inversion to use our own images to train for those of us that are far less program savvy!
With the newest version, aiimages already sets the right values to run on the GPU. (It only won't work on the 1650, because it doesn't have enough VRAM for the fix. :( )
Not with my experience. I rather accidently created some nsfw stuff just by adding the word "sexy" or so. Try adding more words describing what you want?
e.g;I selected an input image, then deleted that input image from the program's folder.(using windows explorer)
then all the pictures I wanted to reprocess took hundreds of seconds, the program went into an infinite loop and the rendering was never finished. gpu usage is %1, cpu usage is %1. vram and ram is totally empty.
You need to fix the coding. There are dozens of ways to put the program in an infinite loop. CPU and gpu usage stays at 1% during this process. this has nothing to do with the system specs, the codes are broken and there are infinite loops.
By changing the input image/delete input image/ playing around with reload a bit, you can loop the program endlessly and then it won't fix.When you open it again, the result is the same.the program becomes useless.
Hello! First of all, I wanna thank you for your work, It's great! I reach out, because I'm facing an issue. The software worked great until yesterday, but today it is stuck in "Processing..." I opened the debug window and it says: "NullReferenceException: Object reference not set to an instance of an object" I deleted the folder and copied it againt, but no luck so far. Do you have any tip? Thank you!
Thanks for reaching out! :) Can you open the options - > open settings folder button - > send me the player file on Discord (Sunija#6598)? :)
Also, if you close aiimages, delete everything in the settings folder and run it again, the bug should disappear (but you won't see your old pictures in the tool again :( ).
If others are curious you can use any image you want in this software if you just rename another png to another png that was created by the software in the outputs folder, and restart the software - it will show you the new images to then inpaint. It also doesn't care about the previous images file size or resolution. Make sure you put clothes on any undressed women you come across since only a sinner would use it to do the opposite.
Hopefully we can get a feature to replace these files instead of having to rename them ourselves though or if someone else has another solution.
Hello, thanks again for the reply. I really want to reiterate how amazing you are at creating software. I use Unity to create VR environments and have spent 1000s of hours as its my daily tool besides Blender and wish I had even 1% of the knowledge you do to create something as impressive as this.
Well, the reason I ask is because I want to "mass import" images, and if you use the load from the PC I am always getting "black" in the section that I use to inpaint, unless I actually physically replace another image by copying another file over with the same file name. I found a trick to mass import 100s of files at once, but an official method to just drag and drop PC files to load into the history to have quick access to make quick inpaints would be really useful since I am using the tool to inpaint hundreds of images and loading individual images straight from the PC always causes the black issue (I can screenshot or share more details if others aren't having this issue). I tried with dozens of images using the load from box, and is always black when I try to use (not created from software, just images on the HD, works fine if I use the load image to use one from aiimages).
I also noticed one minor problem and is maybe my PC resolution (4k, using 200% windows scaling) and is a bug, or is no one else has had any reason to create 40+ templates, but it runs off the screen and you can't scroll down / see more of the templates past a certain point.
Even modifying the UI scale in software, and using the scroll bar or do anything it still runs off the page. I have about 60 templates and can't see past the 40 on screen. Could you please look into this? The reason it is blank is because I moved the images out of the folder to take screenshots so them being blank is not an issue.
Thanks for your time (again!) in answering questions and helping us users in the comments by the way. Its really cool to see a developer passionate about making their product better especially since is such a game changer in the technological landscape. Every person I've seen talk about aiimages says is the best GUI and easiest setup and boy does it work...I'm at 10 gig of images created, nothing takes more than like 20 seconds and can just set a prompt, leave the room and come back to see 1000s of new images. I'm going to use aiimages actually to completely texture an environment and models from blender and Unity to see what kind of amazing thing can be created with just prompts and I bet the answer is "better than if I spent hours (days if not months) in photoshop!"
Embeddings please in the next version also if it isn't too much trouble :)
Hello I don't know why but I can't reply to the earlier comment from you so thanks again for the assistance in implementing waifu diffusion. If you could please add the embedding feature so we could train it on images it doesn't have access to so we can add our own art styles/imagery it would be greatly appreciated. I think you have created the best software ever compiled into exe format, congratulations.
Also, is there something I am missing in how to get the img2img work? It doesn't seem to work with outside images. Even if you replace the PNG file with another file it doesn't show the correct output in the software, but then if you go to the outputs folder you can see it is being modified. This is great, because we can use your img2img feature with any image we want but it doesn't properly show the outputs. Maybe I am creating more work for myself but I can't seem to use outside images unless created with the software? Is this intentional? Because i would like to propose a method to simply "replace" files with other png on your hard drive, since that DOES show the proper data when generated, just requires having to rename each file individually to something in the outputs and then overwriting which takes a while to do lots of images.
Did you start it via the desktop shortcut? That one is broken atm. :X Will be fixed with the next patch. For now, just start it by double clicking the executable.
No, I start the app from the folder. It worked once. If I start it again the app loads the weight, the model, continually and it doesn't finish.
I rebooted and it worked again.
I tried to replace the model file with the mine edited one which it's trained for my pictures but the results were distorted, with overexposed colors and not good composed.
Also deleting the folder and extracting the files again gives the same error.
My laptop gpu has 4 Gb vram.
I used the same model on stable-diffusion-webui installed on my pc and it works perfectly loading an image in 35 seconds, also if the GFPGAN feature to fix face causes error Cuda out of memory, maybe because it needs some configurations.
Hello, first off thank you this is absolutely incredible. My first foray into the world of stable diffusion and I can't believe what I am seeing! Is there any way in the future we will have the ability to train it to get better results? It seems they didn't train it anything explicit to make hentai art with. Thank you.
You might want to replace your model with waifu diffusion.
1) Download the waifu diffusion model here: https://github.com/harubaru/waifu-diffusion/blob/main/docs/en/weights/danbooru-7... 2) Open folder aiimages_standalone\stable-diffusion\models\ldm\stable-diffusion-v1 3) Rename your old model file to "model_original" 4) Copy the downloaded waifu diffusion model there and call it "model" 5) Run aiimag.es
There is also something called "embeddings", where you can teach the AI new words (needs 5 images of the word and 1h on a RTX 3090ti). I might add a UI for that later or - if you are really really tech savvy - you can run it via the "stable-diffusion" folder in a command line. https://github.com/invoke-ai/InvokeAI/blob/main/docs/features/TEXTUAL_INVERSION....
Hello, I already have a working Stable Diffusion install after following a Youtube video tutorial, but it's just a command prompt and so it's not very intuitive to use. Therefore, I really want to use this GUI. However, I am having a problem where the output images are all green squares.
This also happened to me on my other Stable Diffusion install, but I found out that if I type "--precision full" after a prompt, it works as expected.
I have a Geforce 1660 Super GPU which apparently causes some problems for Stable Diffusion. But since I was able to find a solution for my other Stable Diffusion install, I wonder if I could do that here too?
In short: is it possible to have "--precision full" run every time I generate an image, so that the output image isn't a blank green square?
Edit: In the settings json, I found "bFullPrecision" which is set to true. Assuming that this is the same thing as "--precision full", but it still doesn't work, well... that is worrisome for me. Is "bFullPrecision" the same as "--precision full"? If not, my above question still stands, but if it is, then would you happen to know of any other solution for a 1660 Super GPU to output images correctly?
For reference, this is the video tutorial I followed to get a working Stable Diffusion install on my PC, that runs on the command line.
And when I type "--precision full", it works perfectly. So Stable Diffusion can indeed work on a 1660 Super, but I don't know exactly how it works.
As you already saw: Aiimag.es already applies full_precision for those graphic cards, but there's a bug in the backend that doesn't make it work (yet). :(
Apparently the 1650 models might not work at all. :(
The fix that makes the 1650 models work needs more VRAM than the 1650 model has. I'll keep you updated if the backend devs find a way to still make it work! <3
Can you open the options - > open settings folder button - > send me the player file on Discord (Sunija#6598)? v1.1.2 should work on your GPU, so maybe there's another issue. :3
Aiimages uses a history file that is saved to "C:\Users\[Your Username]\AppData\LocalLow\sunija\aiimages" to find previous images. Did that one get deleted by any chance?
Also make sure that the outputs folder is called "outputs" and is in the same folder as the executable.
If neither of those work, you can drop the images in the application or click on the empty input image, to load them as inputs. Also you can drop the images in https://www.metadata2go.com/ to get your settings, prompt, seed, etc, to recreate the images (which is tedious, though :/).
History file and outputs folders are where they are supposed to be and the files names are unchanged. Program continues to work fine. For whatever reason I cannot drag and drop images into the application like you mentioned, but I was able to recreate the images by clicking the empty input image and loading them as input image and running them with the influence cranked all the way up.
as far as i'm aware, there aren't. it'd be cool if they added it in the next update, if the user wanted to exclude certain results. thats just a suggestion tho
I've been trying to get this program to work for a few hours now but all I get it to do is generate the same 4 pictures that use the character at the top right of the page. Is there something I'm missing?
That indicates that something is crashing. You could try generating a smaller image first. If you open the options - > open settings folder button - > send me the player file via Discord (Sunija#6598), then I might give you a better answer. :3
Stable diffusion is - by default - without filter. :)
No local installations has a filter, unless somebody puts in the effort to develop one. The filters that are used for websites like dreamstudio.ai are not open source (as far as I know).
a slight bug report, ive notice that everytime i delete an image, the previous image duplicates. So when i delete a bunch of images, i end up with dozens of the same image. and when i try deleting any of the duplicated image, it deletes all of duplicate images including the original.
hey, whenever i try generating anything, the faces are always distorted and horribly messed up. i tried turning up the better face scale and tried clicking on the face icon but the results were always really bad. any idea how i could fix this?
also, you mentioned inpainting as an option and feature but im not quite access this feature.
1) Inpainting Drag the image into the "Input Image", then click the little blue wrench in the top right of the image preview there. Paint over the face, click "apply" and let it run again.
2) More steps Sometimes faces get better with more steps.
3) Face bigger in the picture If the face is really small in the picture, the AI seems to put less effort into it. If you can make a picture where the face takes up more space in the picture, it should work better.
4) Different prompt Some prompts just generate wonky faces. :3 Next to "style" you can press on the template image. There you can select some style templates that usually create non-distorted faces.
You can also post a picture of your setup on Discord, then we can check together if some setting maybe inhibits beautiful faces. :)
thanks for the reply, I have another minor questions though, what's the difference between the grey slider and the white slider for each of the options?
also, theres an image i generated but its sort of cut halfway. How would i change the width of the image and generate the other half of the image as well? i tried dragging the image the input but i cant seem to increase the width resolution and it just stays the same and cut in half. any recommendations? (sorry for asking so many questions, im just fascinated by the ai)
Grey slider: The value that is used for every image, when you click "start". You can think of it as a preview. White slider: The value that is used when you click on the buttons on an image. E.g. you set "Steps slider - grey: 20, white:50". Then the AI will quickly generate loooads of images, but only with 20 steps, so they might have some glitches. When you found a nice image, you click the circular arrow in the hover menu. Then the image is re-done, but this time with 50 steps, to make any glitches disappear.
Not sure if that's what happened for you: If the AI generates "two images in one", there's not much you can do atm, sorry. :/ If outpainting is added in the future, there might be a solution for that.
Thanks for making this - to make it clear - I'm not trying to bash your work - would simply want to use better version of this tool, and this GUI looks promising.
UI tooltips have very vague descriptions IMHO.
- "Steps" - Really unclear TBH. It says grey bar is for previewing. But there is no preview button. Are grey and white bars somehow related? Asking because they share the same space. If both the grey handle and white handle overlap, what happens? Or are they simply unrelated and probably better as separate sliders?
- "Accuracy" - it says default is 7.5 (which it wasn't). It also talks about yellow bar, which isn't visible by default. How does this actually work is unclear. What happens when yellow bar knob is moved right? Or when yellow bar knob is on the left edge?
- Width/height; Really strange UI, why not simply have two sliders, one for width, one for height?
Popup image preview - I find it really annoying that hovering over small thumbnail brings up a preview image size of half the application window. I'd personally expect 2x or 3x size preview. Why not make larger image preview appear if user clicks or double clicks a thumbnail?
Order of render images - I guess this depends, but I'd personally expect left to right, and then top to bottom. Now rendering order seems a bit off.
Palette - what is the purpose of this window? There is no tooltip.
Rendered images - there is no way to delete files? Dragging to trash only removes it from list. No matter what, I usually delete most of the variants, there doesn't seem to be a way to update list of images, if I manually delete the files. Also, it would be better if prompt was a separate file for each image, that way both can be stored or deleted easily. Now all the output datas are stored in one file, making it really hard to store unique results (like drag and drop the prompt and image in some other folder).
Play - could it possible to make it only render one image? Make a separate button to infinitely generate images?
- Steps tooltip: Updated the description, but it's most likely still a bit unclear. :X I hope I can release a quick tutorial video tomorrow.
- Accuracy tooltip: Also reworked, but might also need the tutorial.
- Width/height: I wanted to save space. :D But it's really not very beautiful. I'll try to find a nice solution till next version.
- Giant popup image preview: Fixed. The default preview is now smaller and at a fixed position. If you click the middle of an image, you can toggle to the big preview.
- Render order: It's top->bottom to work a bit like blogs. :3 If it wasn't left->right it might have been buggy (which wouldn't surprise me, but should be fixed now). Maybe I'll add an option to make it bottom->top.
- Palette misses tooltip: Fixed.
- Deletion: Is a lot easier now. It still only puts them in a trash folder instead of deleting them right away (until I can be sure that nobody deletes something accidentally), but you can just delete this folder from time to time then. Prompts are saved in the image's meta info atm, but you'd have to read those via an online tool... so not very handy. I'll think about a solution until next version. :)
- Tooltip boxes text bad aligned: Fixed.
- Rendered images don't show up until dragging: Fixed.
- UI dies when resizing: Fixed.
- Unclear UI: Semi-fixed. You can disable the background texture in the settings. I'll work on the rest for the next version.
Again: Thank you a lot for the feedback! <3 Helped a lot for the new release.
Some comments; - Delete mode moves renders to trashcan folder, but dropping renders into trashcan doesn't. Is this intentional?
- Main rendered images area already works better, but deleting items / resizing windows (not sure what I did) makes the list look like a jagged array. Some rows are full, others have some items. Is this intentional? Could it simply be a one continuous list, like with Grid Layout Group? I've done several UIs for my own little projects and it works great for lists like this.
- "Prompts are saved in the image's meta" - Great! BTW - simply drop your png to any text editor (like notepad++), and you'll see your stored prompt near the beginning of the file, even though most of the other stuff will be "garbage". Another way is to use some popular image viewer/file browser like XnView MP, it can show you the metadata, from Preview Image panel's Info tab, ExifTool (non-MP version doesn't have this feature AFAIK).
BTW UI has some issues. Seems like it is made with Unity and UI has several typical Unity UI issues.
- Tooltip boxes have texts badly aligned, text goes outside of bounds in left and down directions. - Rendered images don't show up, I had to drag corner of the window to make it smaller, only then rendered images appeared. I noticed that created files appeared in rendering folder and started to wonder what is going on because I didn't see anything in the UI. - The UI window can be sized dynamically, but UI elements haven't been anchored properly, some elements don't scale properly, some start to overlap each other, when window is scaled. - The main list of rendered images has issues with image grid going out of view if window is resized. I don't know if you used uGUI or the new UI Toolkit, but at least with uGUI it is fixable. - UI could use more readable and accessible look, now it is very dark and there are no clear borders between areas. At least make it so that contrast between UI and background can be seen properly. I don't see any use for textured boxes TBH, a simple, clean and easy to navigate UI would suffice.
This looks really interesting, I tried to use another SD GUI (NMKD) but kept running into problems with Python or something. I'm downloading this now (it's taking a very long time - more links/hosting on itch.io would be nice); I'll update here after I successfully test it with my 4GB VRAM Video Card and 12GB System RAM.
This GUI has the potential to be an excellent and simple way to take advantage of some of Stable Diffusion's best features.
Horrible download speeds; after several hours of downloading, the zip file was apparently corrupted. I'm giving it one more shot and downloading again. (Is it too large to host on itch.io?)
Thank you so much for your reply and assistance, Sunija.
I finally got it downloaded properly and it runs very well overall; I appreciate how easy it is to start playing with as soon as it is installed. The tooltips and other features are a great touch.
I'll definitely follow your project and recommend when I'm able.
I'm looking forward to some improvements to the interface and the planned inpainting feature.
← Return to tool
Comments
Log in with itch.io to leave a comment.
Path contain space. Try somewhere like C:\Downloads\ or D:\Downloads\ or D:\StableD. No space in path.
Why only 150 steps max? I used to do 250/300 steps with another older version and results got advantage of it
You can increase the maximum step count by...
1) Options -> open settings folder
2) Close aiimages.
3) Open settings.json with Editor. Edit the line ""iMaxStepCount": 150,". Save it.
4) Open aiiamges.
Sorry that there's not easier way yet. :( Will come in the future. <3
Thank you for your reply. Can I also ask if there is a way to choose a custom output folder different from the default one? I wanted to use the software in an M2 and output it on a external HD
Will you be updating to 1.5 Stable Diffusion that released two days ago?
Yep, next update will have it.
But you can already use the 1.5 model. Just go to options -> add custom model -> follow the steps there. :)
Eager for your update dude. Your app is by far the most enjoyable way I've found to use this tech locally on windows. Thanks for all you are doing!
Still can't get it to work with my 1660TI despite the settings saying it should work .
Thanks for reaching out! :) The 1660TI should work.
Can you join the Discord (or write me directly to Sunija#6598) and send me "open the options -> open settings folder button -> the player file"? :)
Mono path[0] = 'C:/Users/User/Downloads/aiimages-win/aiimages_Data/Managed'
Mono config path = 'C:/Users/User/Downloads/aiimages-win/MonoBleedingEdge/etc'
Initialize engine version: 2020.3.16f1 (049d6eca3c44)
[Subsystems] Discovering subsystems at path C:/Users/User/Downloads/aiimages-win/aiimages_Data/UnitySubsystems
GfxDevice: creating device client; threaded=1
Direct3D:
Version: Direct3D 11.0 [level 11.1]
Renderer: NVIDIA GeForce GTX 1660 Ti (ID=0x2191)
Vendor:
VRAM: 5991 MB
Driver: 27.21.14.6231
Begin MonoManager ReloadAssembly
- Completed reload, in 0.179 seconds
D3D11 device created for Microsoft Media Foundation video decoding.
<RI> Initializing input.
<RI> Input initialized.
<RI> Initialized touch support.
UnloadTime: 0.826100 ms
[10/22/2022 1:14:39 AM] Loading settings
[10/22/2022 1:14:39 AM] Loading history.
[10/22/2022 1:14:39 AM] Loading savegame
[10/22/2022 1:14:39 AM] <b>Your graphics card</b>: NVIDIA GeForce GTX 1660 Ti (6 GB)
Should work! <3
Writing: C:
Writing: cd "C:/Users/User/Downloads/aiimages-win/aiimages_Data/../stable-diffusion"
Writing: activate ldm
Writing: set TRANSFORMERS_CACHE=C:/Users/User/Downloads/aiimages-win/aiimages_Data/../ai_cache/huggingface/transformers
Writing: set TORCH_HOME="C:/Users/User/Downloads/aiimages-win/aiimages_Data/../ai_cache/torch"
Writing: python scripts/dream.py
[10/22/2022 1:14:40 AM] Saving settings.
>>>>>>>> Microsoft Windows [Version 10.0.19044.2130]
>>>>>>>> (c) Microsoft Corporation. All rights reserved.
>>>>>>>> C:\Users\User\Downloads\aiimages-win>call C:/Users/User/Downloads/aiimages-win/aiimages_Data/../env/Scripts/activate.bat
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win>C:
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win>cd "C:/Users/User/Downloads/aiimages-win/aiimages_Data/../stable-diffusion"
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win\stable-diffusion>activate ldm
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win\stable-diffusion>set "TRANSFORMERS_CACHE=C:/Users/User/Downloads/aiimages-win/aiimages_Data/../ai_cache/huggingface/transformers"
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win\stable-diffusion>set "TORCH_HOME=C:/Users/User/Downloads/aiimages-win/aiimages_Data/../ai_cache/torch"
>>>>>>>> (env) C:\Users\User\Downloads\aiimages-win\stable-diffusion>python scripts/dream.py -o "C:/Users/User/Downloads/aiimages-win/aiimages_Data/../outputs"
I've been using this thing non-stop like 5+ hours a day. It just is so much fun.
I've spent the past day or so reading up on the embeddings/textual inversion and google colab/Dreambooth options as well to train the stable diffusion model more, but it seems every option becomes difficult to various reasons - and the other options require you to pay to run a server farm to train the model for you.
Is there any chance you are working on or can put together a UI so we can train and create our own model.ckpt file directly in the software? I've looked at other options but require installing Ubuntu in Windows, or having to go through all that difficult Python scripts when we just want to be able to drag and drop images, click go and come back to a sweet new file.
Please, if you could just include a simple UI it would mean the world to myself and so many others. Your option to generate unlimited images makes it far better than the other stable diffusion UIs.
Awesome to hear that you like it so much! <3
I think you cannot train an own model.ckpt on consumer hardware. :( And it needs about 40k images with labels. But there are two things I might add.
A) Embeddings. You can find those in the documentation for invokeai. Basically: Give the ai "5 images + 1 tag + 1h training on a RTX 3090ti = ai learned new word". I could create a UI for that already.
B) Dreambooth. I don't understand it yet, but it's kinda "embeddings but better", if I understood it correctly. It is not implemented in the backend yet, but it will be implemented. I'm kinda hyped for that, because it allows to create the same person in different poses. <3
Pretty impressed by what it can spit down with a nice prompt and some fine tuning. It can even manage some pretty face and hands instead of the distorted monstruosity that i saw in some other browser based generator.
It seem to get stuck on an infinite loading while i try to upscale or repeat an already upscaled / repeated image, but still for a generous price tag of free is one of the best tool that i've ever seen
:O That's actually a really impressive hand.
Would be cool if you could poke me on discord (either on the server, or directly Sunija#6598) so we can try to fix the infinite loading. <3
HI i just downloaded this and it said that the module took to long if you could help me out that would be helpful
(and if you need my pc specs here they are)
ram :16
CPU: Ryzen 5000 series
gpu:1650
Sadly, it won't work with the gtx 1650. :(
The GTX 1650 has two problems that make it the only Nvidia card with sufficient RAM to not work. :(
all right , thanks
Thank you. All other functions work normally. Only in the inpaint, if you want to modify the image partially, it will be stuck all the time. Why? （3060ti）
Thanks for reaching out! :)
Would be cool if you could poke me on discord (either on the server, or directly Sunija#6598) so we can try to fix that. <3
recently ive seen some rlly cool ai animation videos that show a bunch of ai generated images together like an animated video using stable diffusion and I was just curious if its possible to do with this?
aiimag.es cannot do videos (yet?) and maybe I won't have the time to add them. :(
But if you (or somebody else) finds a cool tool to do videos, feel free to post it here. <3
i asked around and there's something called Deforum Stable Diffusion: https://github.com/HelixNGC7293/DeforumStableDiffusionLocal
i have not clue how to use it but some ppl might find it cool
would greatly appreciate some kind of integration or pipeline for Textual Inversion to use our own images to train for those of us that are far less program savvy!
1660ti may be worked by “——medvram”or“——lowvram” --precision full --no-half
Thanks for the help! :)
With the newest version, aiimages already sets the right values to run on the GPU. (It only won't work on the 1650, because it doesn't have enough VRAM for the fix. :( )
where can I get the newest version?
Newest version is above the one that only says "aiimag.es download". :)
yeah，I saw it，Version 5 14 days ago，and still doesnt work。。
First, thanks for sharing this fantastic tool. It's incredible to create cool variations of self-painted images :)
Is there a way to also extend images with new content like in DallE? :)
Not yet, might come with the next update. :)
(The backend already supports it, I'll just have to write the UI for it. :X)
how to get adult contents，i try some nsfw words but can't work.is there some safe check just inside and invisible?
Not with my experience. I rather accidently created some nsfw stuff just by adding the word "sexy" or so. Try adding more words describing what you want?
also this model takes a Finnneeee time to load
Can you make 512 by 512 images with a 4gb of vram?
Should be possible in general. Did the model finish loading for you? :3
e.g; I selected an input image, then deleted that input image from the program's folder. (using windows explorer)
then all the pictures I wanted to reprocess took hundreds of seconds, the program went into an infinite loop and the rendering was never finished. gpu usage is %1, cpu usage is %1. vram and ram is totally empty.
You need to fix the coding. There are dozens of ways to put the program in an infinite loop. CPU and gpu usage stays at 1% during this process. this has nothing to do with the system specs, the codes are broken and there are infinite loops.
By changing the input image/delete input image/ playing around with reload a bit, you can loop the program endlessly and then it won't fix. When you open it again, the result is the same. the program becomes useless.
nice effort but extremely buggy, need a hotfixes.
anyway thanks for effort.
Hello! First of all, I wanna thank you for your work, It's great!
I reach out, because I'm facing an issue. The software worked great until yesterday, but today it is stuck in "Processing..." I opened the debug window and it says: "NullReferenceException: Object reference not set to an instance of an object" I deleted the folder and copied it againt, but no luck so far. Do you have any tip? Thank you!
Thanks for reaching out! :)
Can you open the options - > open settings folder button - > send me the player file on Discord (Sunija#6598)? :)
Also, if you close aiimages, delete everything in the settings folder and run it again, the bug should disappear (but you won't see your old pictures in the tool again :( ).
If others are curious you can use any image you want in this software if you just rename another png to another png that was created by the software in the outputs folder, and restart the software - it will show you the new images to then inpaint. It also doesn't care about the previous images file size or resolution. Make sure you put clothes on any undressed women you come across since only a sinner would use it to do the opposite.
Hopefully we can get a feature to replace these files instead of having to rename them ourselves though or if someone else has another solution.
In the new version, you can just drag'n'drop an image on the tool. :)
In the old version, you can click on the input image to load one from your PC.
Hello, thanks again for the reply. I really want to reiterate how amazing you are at creating software. I use Unity to create VR environments and have spent 1000s of hours as its my daily tool besides Blender and wish I had even 1% of the knowledge you do to create something as impressive as this.
Well, the reason I ask is because I want to "mass import" images, and if you use the load from the PC I am always getting "black" in the section that I use to inpaint, unless I actually physically replace another image by copying another file over with the same file name. I found a trick to mass import 100s of files at once, but an official method to just drag and drop PC files to load into the history to have quick access to make quick inpaints would be really useful since I am using the tool to inpaint hundreds of images and loading individual images straight from the PC always causes the black issue (I can screenshot or share more details if others aren't having this issue). I tried with dozens of images using the load from box, and is always black when I try to use (not created from software, just images on the HD, works fine if I use the load image to use one from aiimages).
I also noticed one minor problem and is maybe my PC resolution (4k, using 200% windows scaling) and is a bug, or is no one else has had any reason to create 40+ templates, but it runs off the screen and you can't scroll down / see more of the templates past a certain point.
Even modifying the UI scale in software, and using the scroll bar or do anything it still runs off the page. I have about 60 templates and can't see past the 40 on screen. Could you please look into this? The reason it is blank is because I moved the images out of the folder to take screenshots so them being blank is not an issue.
Thanks for your time (again!) in answering questions and helping us users in the comments by the way. Its really cool to see a developer passionate about making their product better especially since is such a game changer in the technological landscape. Every person I've seen talk about aiimages says is the best GUI and easiest setup and boy does it work...I'm at 10 gig of images created, nothing takes more than like 20 seconds and can just set a prompt, leave the room and come back to see 1000s of new images. I'm going to use aiimages actually to completely texture an environment and models from blender and Unity to see what kind of amazing thing can be created with just prompts and I bet the answer is "better than if I spent hours (days if not months) in photoshop!"
Embeddings please in the next version also if it isn't too much trouble :)
Hello I don't know why but I can't reply to the earlier comment from you so thanks again for the assistance in implementing waifu diffusion. If you could please add the embedding feature so we could train it on images it doesn't have access to so we can add our own art styles/imagery it would be greatly appreciated. I think you have created the best software ever compiled into exe format, congratulations.
Also, is there something I am missing in how to get the img2img work? It doesn't seem to work with outside images. Even if you replace the PNG file with another file it doesn't show the correct output in the software, but then if you go to the outputs folder you can see it is being modified. This is great, because we can use your img2img feature with any image we want but it doesn't properly show the outputs. Maybe I am creating more work for myself but I can't seem to use outside images unless created with the software? Is this intentional? Because i would like to propose a method to simply "replace" files with other png on your hard drive, since that DOES show the proper data when generated, just requires having to rename each file individually to something in the outputs and then overwriting which takes a while to do lots of images.
Anyone else getting the "failed to load mono" error when running the exe?
Thanks for reaching out!
1) Are you on a Windows PC?
2) Did you install it to a very deep folder? (E.g. C:/a_very_long_foldername/and_another_one/aiimages/)
If 2 is the case, copy it to a higher folder (E.g. C:/Data/) and unzip it there again. Or use the itch.io/app to install aiimages. :)
Only works for Nvidia graphics?
Yep, atm it will sadly only work with Nvidia cards. :(
AMD might follow with later patches.
Ah okey :) Thanks for the Info
First time I open aiimages the model gets loaded, if I close the application and load it again it loads continually and doesn't finish
Did you start it via the desktop shortcut?
That one is broken atm. :X Will be fixed with the next patch. For now, just start it by double clicking the executable.
No, I start the app from the folder. It worked once. If I start it again the app loads the weight, the model, continually and it doesn't finish.
I rebooted and it worked again.
I tried to replace the model file with the mine edited one which it's trained for my pictures but the results were distorted, with overexposed colors and not good composed.
Also deleting the folder and extracting the files again gives the same error.
My laptop gpu has 4 Gb vram.
I used the same model on stable-diffusion-webui installed on my pc and it works perfectly loading an image in 35 seconds, also if the GFPGAN feature to fix face causes error Cuda out of memory, maybe because it needs some configurations.
i have RX 570 4gb found or not?
The RX 570 won't work (yet), because it is an AMD card, not a Nvidia one. :(
AMD cards might work with future updates.
i see and is very sad :'v
Greatly appreciated! Love the 90s style UI!
Keep up the awesome work!
Hello, first off thank you this is absolutely incredible. My first foray into the world of stable diffusion and I can't believe what I am seeing! Is there any way in the future we will have the ability to train it to get better results? It seems they didn't train it anything explicit to make hentai art with. Thank you.
You might want to replace your model with waifu diffusion.
1) Download the waifu diffusion model here: https://github.com/harubaru/waifu-diffusion/blob/main/docs/en/weights/danbooru-7...
2) Open folder aiimages_standalone\stable-diffusion\models\ldm\stable-diffusion-v1
3) Rename your old model file to "model_original"
4) Copy the downloaded waifu diffusion model there and call it "model"
5) Run aiimag.es
There is also something called "embeddings", where you can teach the AI new words (needs 5 images of the word and 1h on a RTX 3090ti). I might add a UI for that later or - if you are really really tech savvy - you can run it via the "stable-diffusion" folder in a command line. https://github.com/invoke-ai/InvokeAI/blob/main/docs/features/TEXTUAL_INVERSION....
does it still work with the new version of Stable Diffusion?
Any plans on making a Linux build for this?
I'd love to!
But I don't have a Linux system to test the (necessary but few) changes on, yet. :/
Well if you need testing on linux i can be of service. Dont know anything about stable diffusion but running ubuntu and unity on it.
Hello, I already have a working Stable Diffusion install after following a Youtube video tutorial, but it's just a command prompt and so it's not very intuitive to use. Therefore, I really want to use this GUI. However, I am having a problem where the output images are all green squares.
This also happened to me on my other Stable Diffusion install, but I found out that if I type "--precision full" after a prompt, it works as expected.
I have a Geforce 1660 Super GPU which apparently causes some problems for Stable Diffusion. But since I was able to find a solution for my other Stable Diffusion install, I wonder if I could do that here too?
In short: is it possible to have "--precision full" run every time I generate an image, so that the output image isn't a blank green square?
Edit: In the settings json, I found "bFullPrecision" which is set to true. Assuming that this is the same thing as "--precision full", but it still doesn't work, well... that is worrisome for me. Is "bFullPrecision" the same as "--precision full"? If not, my above question still stands, but if it is, then would you happen to know of any other solution for a 1660 Super GPU to output images correctly?
For reference, this is the video tutorial I followed to get a working Stable Diffusion install on my PC, that runs on the command line.
And when I type "--precision full", it works perfectly. So Stable Diffusion can indeed work on a 1660 Super, but I don't know exactly how it works.Thanks for reaching out!
As you already saw: Aiimag.es already applies full_precision for those graphic cards, but there's a bug in the backend that doesn't make it work (yet). :(
I hope I can fix it by today/tomorrow!
and 1650 super, too, please correct)))
Apparently the 1650 models might not work at all. :(
The fix that makes the 1650 models work needs more VRAM than the 1650 model has. I'll keep you updated if the backend devs find a way to still make it work! <3
Man, I will be so happy if you fix this. 1660TI - same problems with green cubes, but your tool is awesome!
Is fixed now. :)
Thank you! I'm excited for it to be fixed in the coming days.
Is fixed. :)
my card is 1660TI too.And i got green cube on v1.1.1,but got nothing on v1.1.2 ...
I'm excited for it to be fixed in 1.1.3
love u
Can you open the options - > open settings folder button - > send me the player file on Discord (Sunija#6598)?
v1.1.2 should work on your GPU, so maybe there's another issue. :3
dude,my card is 1660TI too!did have fixed this problem?
Should be fixed in the newest version. :)
Thank you for using aiimages! <3
Aiimages uses a history file that is saved to "C:\Users\[Your Username]\AppData\LocalLow\sunija\aiimages" to find previous images. Did that one get deleted by any chance?
Also make sure that the outputs folder is called "outputs" and is in the same folder as the executable.
If neither of those work, you can drop the images in the application or click on the empty input image, to load them as inputs. Also you can drop the images in https://www.metadata2go.com/ to get your settings, prompt, seed, etc, to recreate the images (which is tedious, though :/).
History file and outputs folders are where they are supposed to be and the files names are unchanged. Program continues to work fine. For whatever reason I cannot drag and drop images into the application like you mentioned, but I was able to recreate the images by clicking the empty input image and loading them as input image and running them with the influence cranked all the way up.
Your reply was a big help, thanks!
are there negative prompts?
as far as i'm aware, there aren't. it'd be cool if they added it in the next update, if the user wanted to exclude certain results. thats just a suggestion tho
The new update (v1.1.2) added negative prompts. :)
I've been trying to get this program to work for a few hours now but all I get it to do is generate the same 4 pictures that use the character at the top right of the page. Is there something I'm missing?
Thanks for reaching out!
That indicates that something is crashing. You could try generating a smaller image first.
If you open the options - > open settings folder button - > send me the player file via Discord (Sunija#6598), then I might give you a better answer. :3
Hi! Quick question. Does this have any sort of censorship filter on it?
No, there is no filter.
sus
Stable diffusion is - by default - without filter. :)
No local installations has a filter, unless somebody puts in the effort to develop one. The filters that are used for websites like dreamstudio.ai are not open source (as far as I know).
heeeyyy i didnt know how else to message you but im doing well
Oh! Hey! It's good to hear from you! Annoying itch doesn't have DMs, but I'm glad you're doing well. Apologies if it came across as prying!
naaah u good !
oh jesus I think I'm actually gonna make an RPG in this. The challenge: everything used has to be the first draft.
a slight bug report, ive notice that everytime i delete an image, the previous image duplicates. So when i delete a bunch of images, i end up with dozens of the same image. and when i try deleting any of the duplicated image, it deletes all of duplicate images including the original.
Thanks for the report! :)
I'll try to fix that with the next version.
really amazing gui! question though, if i update to the new version, will i retain my saved prompts and styles? thanks!
If you download it manually, copy the "input" and "output" folder to your new installation. :)
If you update via itch, everything should be fine... But making a securty backup of you input/output folders won't hurt. :X
hey, whenever i try generating anything, the faces are always distorted and horribly messed up. i tried turning up the better face scale and tried clicking on the face icon but the results were always really bad. any idea how i could fix this?
also, you mentioned inpainting as an option and feature but im not quite access this feature.
Hi, thanks for reaching out! :)
There are multiple ways to improve faces:
1) Inpainting
Drag the image into the "Input Image", then click the little blue wrench in the top right of the image preview there. Paint over the face, click "apply" and let it run again.
2) More steps
Sometimes faces get better with more steps.
3) Face bigger in the picture
If the face is really small in the picture, the AI seems to put less effort into it. If you can make a picture where the face takes up more space in the picture, it should work better.
4) Different prompt
Some prompts just generate wonky faces. :3 Next to "style" you can press on the template image. There you can select some style templates that usually create non-distorted faces.
You can also post a picture of your setup on Discord, then we can check together if some setting maybe inhibits beautiful faces. :)
thanks for the reply, I have another minor questions though, what's the difference between the grey slider and the white slider for each of the options?
also, theres an image i generated but its sort of cut halfway. How would i change the width of the image and generate the other half of the image as well? i tried dragging the image the input but i cant seem to increase the width resolution and it just stays the same and cut in half. any recommendations? (sorry for asking so many questions, im just fascinated by the ai)
No shame in asking questions. <3
Grey slider: The value that is used for every image, when you click "start". You can think of it as a preview.
White slider: The value that is used when you click on the buttons on an image.
E.g. you set "Steps slider - grey: 20, white:50". Then the AI will quickly generate loooads of images, but only with 20 steps, so they might have some glitches. When you found a nice image, you click the circular arrow in the hover menu. Then the image is re-done, but this time with 50 steps, to make any glitches disappear.
Not sure if that's what happened for you:
If the AI generates "two images in one", there's not much you can do atm, sorry. :/ If outpainting is added in the future, there might be a solution for that.
The new update looks awesome; any chance you could add a update path that doesn't require downloading the entire set of files again?
If you use the itch.io downloader, it should already work like that, right? (Did it work? :X)
Otherwise: Uploading patches would be great, I'll just have to look into it. :) Might take a while.
Thanks for the suggestion!
I've never used the itch.io downloader; I'll look into that. Thanks for the reply.
Lovely, but is there any chance for a version working on AMD Gpu's?
I hope so! :)
There is a discussion to make it run on AMD for the backend that I'm using, but not much progress yet. :/
https://github.com/lstein/stable-diffusion/discussions/407
Thanks for making this - to make it clear - I'm not trying to bash your work - would simply want to use better version of this tool, and this GUI looks promising.
UI tooltips have very vague descriptions IMHO.
- "Steps" - Really unclear TBH. It says grey bar is for previewing. But there is no preview button. Are grey and white bars somehow related? Asking because they share the same space. If both the grey handle and white handle overlap, what happens? Or are they simply unrelated and probably better as separate sliders?
- "Accuracy" - it says default is 7.5 (which it wasn't). It also talks about yellow bar, which isn't visible by default. How does this actually work is unclear. What happens when yellow bar knob is moved right? Or when yellow bar knob is on the left edge?
- Width/height; Really strange UI, why not simply have two sliders, one for width, one for height?
Popup image preview - I find it really annoying that hovering over small thumbnail brings up a preview image size of half the application window. I'd personally expect 2x or 3x size preview. Why not make larger image preview appear if user clicks or double clicks a thumbnail?
Order of render images - I guess this depends, but I'd personally expect left to right, and then top to bottom. Now rendering order seems a bit off.
Palette - what is the purpose of this window? There is no tooltip.
Rendered images - there is no way to delete files? Dragging to trash only removes it from list. No matter what, I usually delete most of the variants, there doesn't seem to be a way to update list of images, if I manually delete the files. Also, it would be better if prompt was a separate file for each image, that way both can be stored or deleted easily. Now all the output datas are stored in one file, making it really hard to store unique results (like drag and drop the prompt and image in some other folder).
Play - could it possible to make it only render one image? Make a separate button to infinitely generate images?
Thanks for the detailed feedback! <3
I just updated the version, so now...
- Steps tooltip: Updated the description, but it's most likely still a bit unclear. :X I hope I can release a quick tutorial video tomorrow.
- Accuracy tooltip: Also reworked, but might also need the tutorial.
- Width/height: I wanted to save space. :D But it's really not very beautiful. I'll try to find a nice solution till next version.
- Giant popup image preview: Fixed. The default preview is now smaller and at a fixed position. If you click the middle of an image, you can toggle to the big preview.
- Render order: It's top->bottom to work a bit like blogs. :3 If it wasn't left->right it might have been buggy (which wouldn't surprise me, but should be fixed now). Maybe I'll add an option to make it bottom->top.
- Palette misses tooltip: Fixed.
- Deletion: Is a lot easier now. It still only puts them in a trash folder instead of deleting them right away (until I can be sure that nobody deletes something accidentally), but you can just delete this folder from time to time then. Prompts are saved in the image's meta info atm, but you'd have to read those via an online tool... so not very handy. I'll think about a solution until next version. :)
- Tooltip boxes text bad aligned: Fixed.
- Rendered images don't show up until dragging: Fixed.
- UI dies when resizing: Fixed.
- Unclear UI: Semi-fixed. You can disable the background texture in the settings. I'll work on the rest for the next version.
Again: Thank you a lot for the feedback! <3 Helped a lot for the new release.
Good job! Already tried the new version quickly.
Some comments;
- Delete mode moves renders to trashcan folder, but dropping renders into trashcan doesn't. Is this intentional?
- Main rendered images area already works better, but deleting items / resizing windows (not sure what I did) makes the list look like a jagged array. Some rows are full, others have some items. Is this intentional? Could it simply be a one continuous list, like with Grid Layout Group? I've done several UIs for my own little projects and it works great for lists like this.
- "Prompts are saved in the image's meta" - Great! BTW - simply drop your png to any text editor (like notepad++), and you'll see your stored prompt near the beginning of the file, even though most of the other stuff will be "garbage". Another way is to use some popular image viewer/file browser like XnView MP, it can show you the metadata, from Preview Image panel's Info tab, ExifTool (non-MP version doesn't have this feature AFAIK).
Thanks, looks interesting!
BTW UI has some issues. Seems like it is made with Unity and UI has several typical Unity UI issues.
- Tooltip boxes have texts badly aligned, text goes outside of bounds in left and down directions.
- Rendered images don't show up, I had to drag corner of the window to make it smaller, only then rendered images appeared. I noticed that created files appeared in rendering folder and started to wonder what is going on because I didn't see anything in the UI.
- The UI window can be sized dynamically, but UI elements haven't been anchored properly, some elements don't scale properly, some start to overlap each other, when window is scaled.
- The main list of rendered images has issues with image grid going out of view if window is resized. I don't know if you used uGUI or the new UI Toolkit, but at least with uGUI it is fixable.
- UI could use more readable and accessible look, now it is very dark and there are no clear borders between areas. At least make it so that contrast between UI and background can be seen properly. I don't see any use for textured boxes TBH, a simple, clean and easy to navigate UI would suffice.
This looks really interesting, I tried to use another SD GUI (NMKD) but kept running into problems with Python or something. I'm downloading this now (it's taking a very long time - more links/hosting on itch.io would be nice); I'll update here after I successfully test it with my 4GB VRAM Video Card and 12GB System RAM.
This GUI has the potential to be an excellent and simple way to take advantage of some of Stable Diffusion's best features.
Horrible download speeds; after several hours of downloading, the zip file was apparently corrupted. I'm giving it one more shot and downloading again. (Is it too large to host on itch.io?)
Thanks for reaching out!
Yeah, itch has a limit of 4 GB. :/ I'll have to check if I can throw out enough to bring it below that limit.Edit: I lied. Itch allows larger uploads if I use their butler tool. Uploading it right now. <3
For me, the download runs at 60 MB/s, but maybe they are lower if you are outside of Europe.
Thank you so much for your reply and assistance, Sunija.
I finally got it downloaded properly and it runs very well overall; I appreciate how easy it is to start playing with as soon as it is installed. The tooltips and other features are a great touch.
I'll definitely follow your project and recommend when I'm able.
I'm looking forward to some improvements to the interface and the planned inpainting feature.
Best of luck on your project!