BEST Automatic 1111 Tricks from my Community!!!! You need to watch this!
Вставка
- Опубліковано 12 вер 2024
- 10 amazing TRICKS for Automatic 1111 to get the most out of it. Quick Access to Clip Skip and VEA loading. Load your last Settings or your SEED with one Click. Save your Prompts and Style Presets. Load the Settings and Prompts from ANY old image made with A1111. Save on Disc Space with Mklink to virtually link to your models with a Link Shell Extension. Set your AI images from PNG to JPG for smaller Files. And more Tips and Tricks to make your life easier
#### Links from the Video ####
CLIP and VEA Settings: docs.google.co...
Link Shell Extension: schinagl.priv....
Opera GX Download: www.opera.com/gx
Everything Search Tool for Windows: www.voidtools.com
Support my Channel:
/ @oliviosarikas
Subscribe to my Newsletter for FREE: My Newsletter: oliviotutorial...
How to get started with Midjourney: • Midjourney AI - FIRST ...
Midjourney Settings explained: • Midjourney Settings Ex...
Best Midjourney Resources: • 😍 Midjourney BEST Reso...
Make better Midjourney Prompts: • Make BETTER Prompts - ...
My Facebook PHOTOGRAPHY group: / oliviotutorials.superfan
My Affinity Photo Creative Packs: gumroad.com/sa...
My Patreon Page: / sarikas
All my Social Media Accounts: linktr.ee/oliv...
#### Links from the Video ####
CLIP and VEA Settings: docs.google.com/document/d/1f6E2J1_b588n4s-Pvd5UqX-S7DwYF14AVkgtBXEBtuc/edit?usp=sharing
Link Shell Extension: schinagl.priv.at/nt/hardlinkshellext/linkshellextension.html
Opera GX Download: www.opera.com/gx
Everything Search Tool for Windows: www.voidtools.com
Please don't forget the VAE explanation video 😊
They ruined the style copy button in 1.6, there's a commit to fix it, but it's not merged yet
Useful trick a lot of people forget to mention: if you right click the generate button you can choose an option to "generate forever" which will just keep generating images until you turn it off by right clicking the button again and choosing "cancel generate forever"
care to explain? why generate 4ever over batch generate
@@Ramirezz11 For my needs: because you may not know HOW long it might take to generate the EXACT image you are going for, if you are looking for something very specific. There are a lot of tricks to make your generation more deliberate, but this entire thing is still basically just a really fancy random number generator, so no amount of effort with the current state of AI can guarantee you get exactly what you want EXCEPT for repeatedly rolling the dice until you get that critical hit.
I certainly find this generate forever to be useful, which is nice because I already knew all the things in the video so I still learned something new.
I think the most useful extension I've installed is Stable-Diffusion-Webui-Civitai-Helper by butaixianran. It's made managing all of my models so much simpler and I no longer have to worry about forgetting trigger words for prompts. It provides the ability to create a sample image for the model, copy the prompts from the sample image directly to your prompt, and also add any trigger words directly to your prompt through a different button. I can add my LoRA's and trigger words now with a couple clicks.
Thank you for this ❤
@@JungzYT No problem!
Since I just had the same issue (switching to Chromium from FF) with dark mode. Two solutions: add the option "--theme dark" to start in dark mode or if it's open add /?__theme=dark to the URL, hit enter and done ;) keep up the great work 👍
Not sure who has done this yet, but you can highlight with control shift arrow left or right, once a word is highlighted while holding shift and ctl you can use the up and down arrows on your keyboard to automatically add bracket weights and the formatting around the selected word that starts at 1.0 and scales forwards and backwards incrementally.
PNG and JPG do not contain the same information. PNG is a loss less image format and therefore contains 100% of the image data. Jpg is a lossy format and throws away a lot of data to reduce the file size. This becomes relevant if you want to upscale the image, where jpg will cause artifacts.
@@JohnSmith-hv6ks img2img*
Personally, I hang onto just the jpg version of x/y/z GRID outputs, since I only look back on them for reference. Everything else should be png of course :)
When he says the jpg has the same information I am pretty sure he was referring to the meta data for recalling the prompt and parameters.
@Sato I think John meant using the same prompt and seed with txt2img
@@mbabuskov yes, if you regenerate with img2img it takes into account the reduced fidelity of the jpg image. thats why u can use photoshop to fix artifacts and replace the image ur using with the edited one.
Auto 1111 already dark mode if you use dark mode on windows..
And you don't need hardlink just put your models anywhere even on external drive then go to (webui-user-bat) click edit and after (Set commandline_args=) add --ckpt-dir and your model path ..
Like this:
Set commandline_args= --ckpt-dir "put here your models path
For example :
Set commandline_args= --ckpt-dir "F:\sd
ew folder"
Then click save and run your auto1111
Don't forget to put (") before and after the path .. as in my example
❤️❤️
you mostly do not need png-info tab. simply drop the image in your "positive prompt" and then use the same button as in 3:03 for populate settings on right places.
Once again, Olivio thank you! You provide such a great service to the community with your clear and concise guides, tips and explanations of this amazing technology!
Thank you very much :)
The seed and settings for best realistic photos. I remember in Mj, there are some seeds like 777, if u nail it, the pictures will tend to be more artistic.
Seeing the cat on the end screen with the heart eyes is the icing on the cake!
Something I found very useful for inpainting is a setting to "save a copy of the greyscale mask". No more redrawing from scratch or going to other apps to perfect that mask.
your zoom / crop is really on point now
Thank you 🙂👍
Thank you Olivio for this awesome and usefull tricks, May IA Bless You.
Thanks, good tips! Though I don't understand the difficulty with Chrome and dark-mode. It works perfectly fine for me on multiple computers. Just add "--theme dark" to the webui-user.bat file. No issues in Chrome. Cheers.
Link Shell Extension is very helpfull,thanks!
you can add " /?__theme=dark " end of your url to use dark mode in chrome
That's great, you can even save it immediatly to a webp image if you want to use it online only!
Also a useful trick: if you don't know the name of the component you want to put at the top of your UI, you can go to the settings, change something temporarily to some value, click apply and you will see the ID of that setting. Copy it, revert your setting, and put that ID into the "quicksettings list" under User Interface tab ;)
The downside is, that you won't be able to plot the custom components, like coherency over 75 tokens, etc.
I recently found the channel and appreciate all the information. Subbed, smashed 👍 and hit 🔔
Another superb video from Olivio! Really really love your content! :D
Great video. Here's my most useful tip. Whenever I get a cool image in txt2img I simply right-click open in new tab. Then I generate some more and do the same. After awhile I can then drag one of the images from the Tabs to the PNG info page and from there back to txt2img or img2img for further work. Or if I'm happy with it I just drag the image from the tab directly to my "work in progress" folder on the desktop.
Nice trick 👌
Thanks for the shell extension tip. I have been using invokeai and recently wanted to try a1111 again but didn't want to have all my models take up twice the space for both.
For anyone wondering, you do the opposite. Start in the invokeai folder, right click and link the folder you have your models in (the hub folder for me), paste it where the stable-diffusion models folder is and rename it appropriately
Hey I don't have the "Hardlink clone" option, any advice ?
@@pn4960 I recently changed up my directory structure (moved stable diffusion to another drive) and I also no longer see the hardlink option. No idea why. I'm gonna look into the other options and hopefully one of those works cause I'm not sure what to do to get the hardlink option back
@@pn4960 Okay so, I created it as a junction instead and it seems to be working fine! I tried to read the download page but it's just so much information I'm unfamiliar with that I don't REALLY know the differences between hardlink and junction and if there's any reason in this specific situation I'd want one over the other. But hardlink isn't showing up anymore for me either so I have no idea.
Hardlink clone only works if you're using the same drive. The reason you're missing that option is because you changed the directory structure. Ideally you'd wanna use Symbolic link which works for both files and directories ( can be used for different drives unlike Hardlink) and is the latest increment in the software
Thanks to you for compiling them and to the great community, these tricks are great!
The best tricks I have seen so fare.
Powerful Tips!! Thanks Community and thanks Olivio
awesome hoodie!
Great video, most useful ever!
Chapters could help!
Thanks!
great tips I used Link Shell Extension for ROMs that are really heavy for the for the disk I have my roms files
Thanks for this information Olivio, very helpful
You can also add a drop down LORA quick selection box with your checkpoint, clip skip and VAE boxes. Just add "sd_lora" with the others. For example in user interface, settings paste --> sd_model_checkpoint, sd_vae, sd_lora,CLIP_stop_at_last_layers and now you will have LORA as well.
Ok, I got it. It has to be written all in lower case as eta_noise_seed_delta
Using lora this way is will messup your lora option. Once you select one lora from the list there is no way to deselect or change value, its a permanent. you can only change the lora from the list. even if you remove the UI selection box the lora will still apply.
Thank you for your videos. They're so useful and your youtube channel is like my SD bible
Thank you so much :)
OLIVIO can you make a video about the best settings showing also prompt and result? thanks!
there are no "best settings". Just like with cooking it depends on what you want to get out of it and what you put into it
My Auto1111 has always been in dark mode and I don't have an extension installed and I'm on chrome. You just change the Gradio theme under settings....
thank you that was useful
Thank you!
For the dark theme on chrome. Put "--theme dark" on your command line in your webui-user bat
jpeg images are better for sharing online
but png images are closer to bit map and keep more pixel data.
PNG is lossless, so it keep all pixel data.
That's a complex workaround to have ComfyUI find the models of A1111. It already has a feature to modify paths, in the extra_model_paths yaml, as documented in the install instructions.
You are right but I believe that comfyui was just an example. Many people use automatic1111 with invokeai or vladmantic for example so sharing the models folder between those can save you hundreds of gigabytes of hard drive space
This is very helpful. Thanks!
Olivio i love your videos and learn for its
5:13 wait comfy ui alreay has path sharing with automatic1111 just add the stable diffusion folder to the example yaml file in the comfy ui folder. but i can use it in other things thanks
Thank you ❤️👍
Thanks for the video. When you saved "Negative 2" I see you 5 buttons under the "Generate" button. I only have the "read generation", "clear prompt "apply all selected styles" where did you get the other 2 buttons?
1:15
You can also add your hypernetworks to that list, as well, if you'd prefer to add them that way. But, kind of redundant with the "Extra Networks" tab available.
i like your negative prompt list ! have you a .txt for this one?
Olivio, first thanks for all the very helpful videos.
I'm sure I've seen you showing how to set the output sub-directories using date. I thought it's in this video. But I just can't find it. Can you help?
Cool Tools
Thank you :)
@@OlivioSarikas before I start training and dataset my models, I have to create 8-15 images of my Original Characters models from different countries
The vae one is very useful. Cheers.
Thank you. I use it often too
Great video!
I dont ustand some of the tec words and slang. I also find the AI programs looks like PC's in the early 90'ies. And that you almost have to be a coder to manage this "prompts" etc. Someone really have to clean up the lack of user friendly Chatgpt. Tutorials for slow learners and or people with english as 2 or 3 language if use isnt improved even more than this attempt. I am going to use it when it has become easier and more intuitive tools added. Thanks for showing the amazing work possible if your a programmer😂
thanks! so helpful
--autolaunch is a commandline argument to launch automatic1111 without needing to click on the link in the cmd!
5:33 --cktp-dir coamdline argument . I'm using it to save the models on an other disk, but could be useful with two UI.
I'm seriously addicted to messing around with stable diffusion 😅
I usually also turn on the add new txt file that accompany each image output. It saves all the output setting (aside of adding it as meta tag inside image). It's easier to access.
I did that in the past, but it get's really confusing and you need to copy everything over by hand again (as far as i know). The PNG info tab is much faster and easier
Is there a way to automatically organize the images i generated by using an SD interrogator or something like that?
I would be very interested in that as my folder has become a total mess.
Or maybe a browser or something that can search just for the prompt metadata i am looking for?
Great tips. thank you.
How did you get the output by date? I can't find that option.
Just was like that after a update. I'm sure it's in settings somewhere
I think it does it automatically. At least it did for me.. Never changed a setting.
@@MaximusProxi ??? Strange. It never did for me and I have installed it multiple times. Hope someone knows where I can select it.
@@lithium534 have you ever usedt the "git pull" command to update your sd version?
Opera also have a built in VPN for free.
U also need to reload UI...
My interface looks a lot different than yours Olivio, I dont have the clip skip right from my checkpoints bar. How do you get it there? I also use Stable diffusion with Automatic 1111. Thanks in advance and for the great vids :) Greetings !
0:53 ...
You can watch the video before posting a comment... It does help a lot...
thanks!
Please, explain to me why you SHOULD use dark mode.
I have enabled the CLIP_stop_at_last_layers option, but it does not matter to what I set it to, the picture always comes out the same (when using the same settings and seed). For some reason Automatic1111 does not take my selection into account. Any ideas?
Comfy already has a way to indicate where a1111 is and take the models from there
Really miss your Affinity videos. Do you think you’ll ever get back with that or have you left that galaxy with AI.. lol.
I am using edge browser, and I don't need an extension for the dark mode.
Thank you very much for the great tricks. I like the Opera GX a lot. However Inside inpaint the color picker seems not to work. Does someone have a workaround?
In there a way to fix the Hires Fix? It is horrible.
You are not using --xformers in the startup parameters, is it for some reason?
Wait a sec....how do you have outputs being saved to a folder with the current date?? That's awesome!
That's because I'm build different 🤣 Na, just kidding. Go to settings -> Save to a directory -> check "save images in subdirectory" and add [date] to "Directory Name Pattern"
@@OlivioSarikas haha truth! Amazing, thank you so much!
3:00 I want to know how to save all of these settings to an external JSON file to load in for later use
I posted similar and now found what I was looking for Model Preset Manager
Does anyone know why my embeding training has started generating random images all of a sudden? It was working fine before & I haven't changed anything. Would reinstalling auto 1111 fix it?
4:12 I use --no-half most of the time because a lot of the models I use need it, because otherwise the console throws errors that say I need to use the no-half argument
Same for me. It's worth having just to avoid those errors.
I wish there was a way to turn it on per model
@@OlivioSarikas That would be a good feature to have.
I had those errors quite frequently a while ago, but I somehow resolved them.. Not really sure how though. I think for me it had to do with using the lora dropdown + ceratain VAEs. I'm not using the lora dropdown anymore and now everything works fine. Not using the --no-half anymore.
@@MaximusProxi do you mean VAE drop-down? loRA's use prompts
Ty for your video! Can you please explain me how to sort folders by date like you ? 9:13
Would it be possible to get a tutorial for the new 1.1 version?
if you want to add anything to quicksettings: change the setting you want to add in a trivial way, then apply. 'The_Misc_Setting' or whatever it is called will appear in text at the top of the settings page informing you that it changed. Then add that text, The_Misc_Setting, or whatever it's called, to the quicksettings field
quick question! can i make lora models with 4gb vram 1050 ti?
Super
Thank you 🙂
太酷辣!
Why use dark mode?
How do you find and install the VAE?
I don't have the "Hardlink" option for the shell extension :/
It is anyway better to use symbolic link on the folder itself, that way you have only one folder you need to manage. Hardlink is in this case the wrong solution.
Why do you recommend to start up with ''Webui-user'' instead of ''launch'' ?
hm... to be honest, i never even saw that there is a lauch.py - does it do something different?
@@OlivioSarikas seems to be a bit faster, but I dont know if there are any differences, thats why I was reaching out
I don't quite get the point of the shell extension. Windows allows you to create symbolic links from an admin command prompt so I have my models in a separate folder and link to it like this: "mklink /D E:\ComfyUI_windows_portable\ComfyUI\models\checkpoints E:\SD_Checkpoints\" and similarly with automatic1111 "mklink /D E:\automatic1111\stable-diffusion-webui\models\Stable-diffusion E:\SD_Checkpoints\"
Yes, but this is easier. Not everyone likes typing commands into a cmd
This is what I did - took all of 1 minute
Hey Olivio, I have recently installed stable diffusion automatic1111 on my Mac m1 pro and it does work however it's extremelyyyy slow. I have added sadtalker extension which just gets stuck on "face renderer" stage after uploading the pic & audio , would really appreciate if you could let me know any possible solution here. Btw, I am on 16gb system
You need to use Draw Things or Mochi. Less customization but it uses the Mac neural engine so it’s blazing fast and easy on Ram. I can generate multiple batches in a minute with an LCM Lora.
opera is your sponsor??????
i always crash when i use hires fix, any idea why? also when it works, sometimes it gives me black rectagle in the left :c
check if there are error messages in your Command Line Windows and ask in the Stable Diffusion Discord what that could be
Hi Olivio, can you help me with this please?... I have rtx 3060 6gb laptop...It is able to generate a batch of 4 images in 15-20 seconds at 512x512. I am ready to wait 5 or 10 minutes too, but want to generate images at 1024x1024. I get the error Out of memory CUDA error. 768x768 is the max resolution at which I don't get this error sometimes.
If you're generating in batches, try setting the Batch Size to 1 and only use the Batch Count. You'll still generate multiple images, but it won't generate them in tandem meaning it won't eat up your VRAM. I have a GTX 1080 and can generate up to 1024x1024 images without a problem, but I can't do batch sizes. I can only increate the batch count.
I dont get the dark mode part. people tell you to get rid of the dark mode in auto1111? why?
leave it as it is, keep your eyes peeled
put Windows to Dark mode
No they tell me to use dark mode in my videos
Just set windows to dark mode and the A1111 UI is dark too.. At least for the Chrome browser, I'm not using anything else really, so I can't tell..
@@OlivioSarikas that makes more sence :-)
Would be interesting to know, why you blocked my account without any reason.
If you mean Facebook, that's my mod team. They do good work. Maybe you didn't read the Community rules. My Facebook group doesn't allow everything
These clip skip comparsion look all the same to me, honestly, where is the improvement?
You need to look closer. Really really close. Like Atomic level inner values look into my soul close. The truth will reveal itself, my friend.
There are only 2 VAEs , novelai and stability , stability had 2 versions, latest one is best
No, there are many VAEs out there. Check the internet
@@OlivioSarikas not really dood, most are copies from nai and stability, maybe some were trained for a bit but theyre mostly from these 2, you can easily check this by comparison and same seed
@@krystiankrysti1396 What about OrangeMix Vae? Also there are versions and they have different strengths. AOM3A2
Features: Oil paintings like style artistic illustrations and stylish background depictions.
AOM3A3
Features: Midpoint of artistic and kawaii. the model has been tuned to combine realistic textures
AOM3A1B
AOM3A1B added. This model is my latest favorite. I recommend it for its moderate realism, moderate brush touch, and moderate LoRA conformity.
Sound to me like they are indeed different. similar, but different.
Yes and no. Some vae's indeed are just renamed version of NAI vae, but most are completely different and will produce different results.
In the saving parameters i've chosen to have " save text information about parameters as chunks to png files" and also " create a text file to every image with generation parameters " like that i'm keeping all the informations on a txt file which it's very low weight
the only problem with ai art is we made the ai models all thirsty, like sometimes you really gotta put in work to get a male with some models which is annoying
true. or getting a average looking person out of most models is pretty much impossible ;)
I have a problem and no solution. Maybe someone of the community can help me, please? The part of the automatic 1111 window where the results are shown, show no more results. Since 3 or 4 days. I did not change anything in the settings. I have to go to my folder for watching the images. And from there I can load an image to PNG info to send an image to img2img. Sure it works but it's a roundabout. Has anyone an idea, what's wrong with my automatic1111?
You could try going into 'Settings', then 'User Interface' and make sure the options from 'Enable full page image viewer' and downwards are ticked. Maybe it's something there?
@@Elwaves2925 It was worth a try, but all options are ticked. But thank you :)
@@PSandmore16 Ah okay. Good luck getting it sorted out.
You need to wait 2-3 seconds before talking at the start of each video.
I feel like these are not tricks but basic usage of the webui. Title is clickbaity and misleading
Jpg is traaaash
I can only select: sd_model_checkpoint. It creates it's own space with an "x" behind it to delete it again. I'm unable to add "CLIP_stop_at_last_layers, sd_vae" behind it.
get dark gradio theme by adding this to address bar at the end of the address, ?__theme=dark
Thank you. They should add that as a setting