Best NEW AI tool? InvokeAI tutorial for Stable diffusion.
Вставка
- Опубліковано 5 жов 2024
- GREAT OUTPAINTING!? In this InvokeAI tutorial we'll check out the latest Stable diffusion ui with great user experience. Will InvokeAI be able to take on automatic1111? The developers of InvokeAI promise stability and a greater focus on the user interface. I just LOVE the outpainting UI.
github.com/inv...
huggingface.co...
huggingface.co...
Support me on Patreon to get access to unique perks!
/ sebastiankamph
Chat with me in our community discord:
/ discord
Ultimate Stable diffusion guide
• Stable diffusion tutor...
Ultimate Animation guide in Stable diffusion
• Stable diffusion anima...
Dreambooth tutorial for Stable diffusion
• Dreambooth tutorial fo...
5 tricks you're not using
• Top 5 Stable diffusion...
Avoid these 7 mistakes
• Don't make these 7 mis...
How to fix live render preview:
• Stable diffusion gui m...
CHAPTERS
0:25 Installation
1:39 Requirements
2:40 More installation
6:01 Using InvokeAI
8:49 Text2image
13:37 Img2img
14:20 Unified canvas
18:54 More features
You can't disappoint me... 😉 great tutorial Sebastian. 🙏 And Happy Holidays to you all Ai Enthusiasts. hohoho 🎅💰
Haha, glad to hear it my friend! And thank you as always. Santa came early this year, eh 🎅🌟 Hope to see you back in Discord whenever you're less busy again!
Wholesome.
SEBASTIAN? forsen E
One really clever thing about the Inpainting feature they have here is the fact that it scales up and then down again depending on your needs. This means that you can zoom in, inpaint something quite detailed and then go back out. I haven't seen that before and when playing around with it I've been able to make some pretty insanely detailed pieces.
Very cool indeed!
Really have my eye in this. Already amazing. The potential is insane. Thanks for making this tutorial.
Also, their install is crazy simple. They're really doing a great job
The biggest problem with Auto1111 is that Gradio won't allow you to use an eraser tool on your mask while inpainting, so you have to get everything right on the first brush stroke. Gradio doesn't allow extending the functionality, so we're basically stuck with unusable inpainting on Auto1111.
Best feature in my opinion is the ability to draw directly on the canvas. I don’t like to go back and forward to my drawing software when I want to do some quick img2img based on a sketch.
Krita has a perfect extension for this from interpause (extension for A1111). It create the transparency mask automatically. You can easily combine a lot of image that way, keeping the best part. It's so strange to me that even quite proficient artist doesn't use it.
It definitely LOOKS friendlier, but looks are just one small thing.
Personally I use prompt matrixes, grids and S/R in Auto1111 very frequently and I don't see an option for that here. Plus extensions like Wildcards.
The moment it reaches the features Auto1111 provides, while keeping the sleek UI, then for sure I'mma trying it out.
I don't think it ever will. Their goal is user experience and stability, so auto1111 will always be the front runner for features when comparing those two.
That's a great, smooth AI! Downloading right now.
This looks really good. Im definitely going to use this today.
Try it, let me know what you think!
thee new interface has great potential for sure, all the tools in one package.
A1111 is gonna have some big competition here. Good to have many options.
Can't believe this is all free, but for some reason, I can easily see InvokeAI start to charge for their services. Great tutorial and intro to this thanks.
They only charge for cloud computation. Charging for an offline app makes no sense these days.
Masking an edge before outpainting it will get you better results. This works in Invoke as well as PaintHua outpainting.
Great tip!
"Masking" means "telling it where the edge is"?
@@dialecticalmonist3405 Its not so much telling it where the edge is but more of a "don't touch these pixels". Masking can be done anywhere in the image not just the edges.
Seems like the future GOAT of the community UIs
I started using this the other day, and I can say it makes generating images a lot easier than automatic's. The biggest thing being the unified canvas. With automatic's interface outpainting is just too cumbersome, having to enable and disable scripts and adjust sizes and settings every time. The unified canvas is amazing and just allows you to outpaint, inpaint or image to image automatically based on what is selected. Feature wise invoke is lacking a bit, but I think given a bit of time they will add a lot more features. I think automatic's interface is a lot better if you want more fine tune control with scripts. If automatic's interface got something like the unified canvas I would probably switch back, but as things are now I have only been using invoke.
I still prefer A1111, but InvokeAI1111 would be the best of all worlds!
It's unfortunate that 2.0 comes with controversial censorship around the fact some people are cringely horrified by the fact that humans have bodies, healthy bodies, and that they can't even differentiate between someone alive or someone dead in their filters, which makes your mom and dad and what did so you're alive... Equal to Ted Bundy and Jeffrey Damer Which is so exaggerated.
What's perturbing is that some people probably prefers violence... To sane human relations, and would censor the latter before the first.
Some people are so cringe that they'd censor renaissance art because it displays not so taboo body parts.
nsfw censorship is disgusting!
@@sher1x165 It depends on the category, that's why NSFW should be separated in two different categories so gruesome materials doesn't mix with simple lack of clothes which isn't shameful nor a taboo.
And the developers are super nice and helpful. Just had to say this.
Ayyy! It's here! Good to see it, Sebastian!
n on that nsfw setting, a man of culture no doubt. Great video. That outpainting is killer, comfyUI generations with invokeAI outpainting, best combination imo
I was excited to try out this new GUI then I saw the whole "don't use this if you have GTX 1650 1660" card. I have a GTX 1660 Super and no wonder why I need to have " --precision full --no-half" in the the .bat file to stop the black square problem.
Awesome content man as always. Thanks!
Thank you for the support 🌟
Just wondering: you twice say a 'write' token is needed, the installer, though, asks for a 'read' token. Can you elaborate on why 'write' should be chosen?
I have a two year old mini pc with maybe half the specs. doesn't even have a gpu but integrated video. and it still kinda works. I can't wait to get a decent pc again
you helped me so much! thanks for all this amazing knowledge!
You're very welcome. Thank you for the kind words 😊
That unified canvas feature looks pretty cool.
It’s not a killer, it actually creates less detailed images.
7:35 If I hadn't already hit like on the video...I would have done it here.
I have 3 Stable Diffusion forks installed on my computer......... AUTOMATIC1111, Stable Diffusion UI, and InvokeAI.
.
The interesting thing is that even when I use the same prompt, try my best to apply the same settings, use the same model...........the outputs from these SD forks are typically quite different looking in style comparatively.
They are not using the exact same configuration. They way they weight your prompt's words is often especially different. The sampler, resolution, steps, cfg scale, clip skip, and many other settings makes a difference too.
thanks for help it works on AMD GPU
In this space "things escalate very quickly"! 😁
For me NovelAi was more slower with text2img.
They also don’t seem to have newest sampling methods.
I am looking forward to their development process because I like their UI better than AUTO1111
Also it crashes very often
Wonderful tutorial. Thank you.
I wish one of them would look at using multiple GPU's. Might put my old mining rig to use.
Thank You!
Thanks for sharing and wow that outpainting tool is really cool, InvokeAI is worth it just for that alone.
Yep, so many users will ignore feature rich alternatives if the user experience is superior.
Thanks Brother for your guidance! I REALLY appreciate your jokes
Thank you for the tutorial. Your channel is amazing.
If I install InvokeAI, will it interfere with Automatic1111 or everything will work normally?
Thank you again
Hope I get this running on Colab. Thanx for the early bird news.
SEBASTIAN, NICE VIDEO!!!
Thank you very much my friend! 😊🌟
I think it's great so far. My only problem is with performance, compared to auto1111 it seems to run so slow and often crashes often for me. I'm running on a 3070, would definitely use Invoke more if it didn't crash so much! Hopefully the performance will be improved with updates.
May be it's not using your GPU,that can happen its running on only cpu...if some script not support for ur GPU
I haven't had it crash at all and I'm using the same card. Try reinstalling it maybe. Hope you get it to run properly, it's a lot of fun!
Truth is I am not able to run invoke ai on nvidia 3060 ti after installation without any error ..its just crash and closing with red ,yellow,green text promts...
I am trying to apply all solutions since 6 day's and now I am tired.
That's pretty strange, I'm running invoke on an old 1050 TI and it works perfectly and faster than automatic. Hasn't crashed once so far.
Thank you so much for all your hard work making these videos. I'm a bit of a noob when it comes to 3D modeling. I'm just curious if stable diffusion can be used to make my 3D models look more realistic, especially around the eyes. I'm not too interested in changing the scenery or the models all that much, but rather, just make them look a little more realistic. I just wanted to know if this is the right program for this? Or, should I start somewhere else first? I'll mainly be focusing on 3D animations and I think I can do that with the use of Krita along with stable diffusion. Any advice would be greatly appreciated.
Will work great for stills at least
Thanks broski
You're welcome my man! 🌟
This tool has an amazing UI, but adding more models is not as easy as in A1111. I hope they fix that soon, because the canvas is the best so far that I have seen.
Great job as always.
Thank you kindly! 😊🌟💎
I haven't tried it myself yet but on the screen it says you need a"read" token, not a "write" one. Though I guess write will also give you read access.
I used a read token and worked
@@andreialcaza just tried it as well.
This is awesome, but I don't think automatic1111's UI is completely useless, it's just not pretty, but it's not a problem either, or a "deal breaker". Stability wise, 1111 it is strong stable as long as you know and understand your equipment capabilities. Great video btw! Good Job! this is awesome content!
automatic 1111 + invoke ai = a revolution for the creation. I have only one issue : the install is on C and i prefer to put nothing more than the OS in it. But anyway i will use invoke for the inpaintang/outpainting most of the time
I easily installed it on my external, fwiw. :)
@@brynbulloch interesting ! You copy/paste the path at the beginning of the install ?
@@brynbulloch and how are you done this ? copy and paste the path where you wanted to put it when it asked you at the beginning of the install ?
@@hatuey6326 yes.
@@brynbulloch Thanks so i will try to uninstall it fromm C to reinstall it on an another drive place
Thank you for the video
Thank you for the comment!
Invoke AI produces images 2-3 times faster than Automatic 1111 on my machine (Ryzen 5 3600x with 8gb 3070ti), but I miss all the extra features of 1111. Invoke's outpainting definately seems better to me, so for now until i figure out the slowdown, or upgrade my gpu, I've been rendering in 1111 and then dragging it into Unified Canvas to outpaint when needed. Not a perfect solution but its been working so far.
Can you do one where you're doing avatars from self-portraits using Invoke AI?
This point. 7:29 :) Best first step ever.
Haha, thank you very much, you absolute superstar you! 😊💎🌟
I like how in depth I can get with Automatic1111, but I do feel it could use a lot of UI cleaning and some UX improvements to streamline the process.
Yes, let's hope this spurs a change in auto1111.
I love this but I get CUDA errors very often
I get cuda errors after OPEN the invoke.bat
Damn you had me in the first half, but i got a 1650 ti in my laptop and i cant swap it out. Feelsbadman.
Hi Sebastian. Great tutorial. I installed invoke this morning and is working fine, with only one issue...when i switch the model to inpainting 1.5, I receive CUDA error. Is there any way how do I fix that?
Great tool. I might switch simply because the outpainting is much more superior to the experimental one in A1111. You can't even choose how exactly you want to outpaint in A1111, only up down left and right lol.
FYI: If you run the installbat and keep getting the message:
Python was not found...download from Microsoft...or disable from Settings > Manage App Execution Aliases
To find this go to Start and type "Manage App Execution Aliases" hit enter, and turn off Pythonexe & Python3exe
Verify by going into Start > CMD and type:
python --version
(returns current version, e.g: Python 3.11.1)
pip --version
(returns current version, e.g: pip 22.3.1 from C:\...)
From there you can run installbat and pick your install folder. 👍
Got stuck for a bit at the installation process, had to deactivate my Antivirus (Kaspersky) to be able to install it properly. Also, don't mess up the HuggingFace token since it remember your token (in cache), if you did, just empty your Windows Cache (in Windows properties of your harddrive) and start over.
This is nice
I think this is the sort of game changer that makes SD better than midjourney
great tutorial! i have a problem! :( ivokeAI installed successfully on my computer, but the invoque.bat file dosnt appear!! How i can fix that?
Same problem here. If you find anything out please let me know.
No negative prompt? I'm a bit of a newbie, but I thought that was a fundamental part of SD. 🤔 So that's just an auto1111 thing?
11:18
It does save all your images by default!
On my computer it's at: C/Users/Louis Gedo/InvokeAI-SRC
In there, it breaks it up into several outputs folders.
You're right. A dev told me this afterwards, I had misunderstood it, the outpainting needs to be saved manually. All generations are saved automatically.
What would be amazing prompt-wise would be the ability to add #ignored text# markers so we could disable parts of the prompts only temporarily, or even having an infinite checkbox array with textboxes where different types of prompts could be inserted... EX: an elephant on a beach ball could become... [x]An elephant with broken ivory tusks and a tutus [O]On a blue, white, yellow and red beach ball made of plastic [O] In a circus with agitated crowd
I wonder if it is possible to use a low res image, stretch it really large, and inpaint it inside InvokeAI? I'll try this right away!
i was hoping this tool will include the feature like auto11111 allowing me to train AI with embeddings and images.
I have a 1650 super and it works well with NMKD Stable Diffusion GUI . I can create 768x768 with full precision and low memory mode at the same time. I've tried Automatic1111 and I could only create 512x512 with CPU, it was slow and horrible.
use this one with old4GB GTX 860 in laptop and running just fine and seem more memory efficient than Automatic1111 because I`m still can doing browsing and something else beside generate images. Something I hardly can do with Automatic ones .
The output seems have more contrast and vibrant color although using default low scale(7.5) . In Automatic 1111 i need to push CFG scale to get same color contrast and vibrance
The installation process cost you around 13- 15GB without any models . Using custom models is possible with editing model YAML texts and with pointing the model path into your Automatic1111 model directories
tytytytyty!!!
no invoke bat ???? any idea ?
no help?
I don't understand why they don't make a dedicated hosting for this so that people who don't have access to GPUs don't have to install it. Is it related to legal issue or something?
how is this vs painthua with automatic?
What's the verdict? It's been 7 months since this video was uploaded. I'm fresh only 2 weeks in and learning.
Tried to install this so many times so far and can never manage to get the local host 9090 page to launch, just get ERR_CONNECTION_REFUSED. I've tried every solution I can find online (that I understand and can follow) and nothing has worked. The only thing I've not tried and I can't believe everyone who seems to have this working is doing this, is to turn off my Firewall. I would love to get this to work, everything seems fine and normal, it just doesn't open. I am curious how everyone I've seen do installation demo's seems to just launch it, not a single person has hit this problem and then overcome it in a video so that I can see or follow what they are doing. Is my computer just set up differently to pretty much everyone else's? It has been a frustrating 7 or 8 hours of standing at my computer trying to give this a go...
If I can't get it to work, what are recommended alternatives?
That sounds like it could be a possible firewall issue alright. Very strange that it would do so locally though. But maybe you're on a very specific setup? I'd try disabling it (can just disconnect from the web temporarily) and see if there's a change.
getting 3 times the information, not working oin my systme here
No negative prompts?!
Yes you need to put them between square brackets [like this]
@@superbaiart1956 🙌🙌🙌🙌🙌🙌🙌Thank you
thats so bad ass yo
Great tutorial. But I installed the most recent version 2.2.5 (successfully) - and only got a blank page in the browser. It's a bug, and I hope they will fix it soon.
i didn't know that i need to know how to code with InvokeAI for example to add models to have to manually editing the file models.yam and Prompt Features
no invoke.bat is fixed with version 2.2.4. released yesterday but before installation you need to install latest version of python.
The secret of great result is negative prompt. I haven't use InvokeAI yet but seems it doesn't have this option
Negative prompts in brackets [ ]
I don't think it will fully replace Automatic1111 -- InvokeAI is more streamline, easiest to use and has cracking outpaining out of the box (also available in a few ways in a1111), but a1111 has unsurpassed flexibility and allows for tinkering where InvokeAI does not -- like setting up upscaler: I personally find SwinIR better than ESRGAN, and this choice is not available here (even worse than that, upscaling via running img2img with SD Upscale script seems to be far superior to the generic Upscale option that also seems to be in use here).
Inpainting/outpaining here is a killer feature though. I'll probably be running both and using IAI for only that.
6:00
The installation didn't work for me though.
And yes I was actually talking with the developers about my problem - they were really nice trying to help me. Big props. But still it doesn't work. (RTX 3090 btw).
Great video, and the UI looks very great indeed, despite it's still missing "everything" that Automatic1111 UI has
Control Net + Lora + hypernetwork + embeddings + extensions + Photoshop Plugin = indispensable tools
Can anyone tell me please what is the difference between Stable Diffusion and InvokeAI? I started using SD a few days ago and I really liked it and I liked the whole topic of AI, now I'm trying to learn more about it) For me, it seems like SD is a some alrorithm of using its models and InvokeAI is a advanced WebUI for SD models, isn't it?
Those are just different front-end. Stable Diffusion is the engine behind the hood if you want.
Given this a try the installer is nice and the UI looks nice but at the moment I'll continue to use AUTOMATIC1111 it's implementation of negative prompts is much better and that alone is reason enough to use it. But I actually like the UI layout more too though maybe in part that's because I'm used to it.
Could you clarify something for me please?
You dont mention the ROCm driver requirement for AMD cards on Windows systems.
Does this mean this only works on Linux?
How's this different from DiffusionBee? Had a chance to compare?
Hey, dumb question: How do I negative prompt with invokeAi? There's no special box for it
Doesn't seem to be working on Mac M1. The installation fails to find Python libraries.
That's a bummer about the 1660 series. I'm mostly mad at Nvidia for screwing that up.
So I put in the "read" for the token because it asked me to. What does that mean?
I never saw a "license agreement" button.
Am I downloading some nerfed version?
*** Awesome, I have no "invoke.bat" I guess that's what "read" permission gives you. Nothing.
why are the out-painted parts worse quality-wise than the starting image? much blurrier and less detailed?
The prob with InvokeAI is they refuse to support 1650 cards, they kind of gave up on it, thinking there was no real reason to continue, so I use automatic.
I'm jumping in when InvokeAI will have deforum in it
What about an equivalent of the highres Fix in Automatic1111 ? Problem (huge) was that it is available only during the processing of new images, but not at all available for img2img or inpainting... :/
I moved to InvokeAI hoping that this would work. But now there doesn't seem to be any option for anything bigger than 512x512. Any idea on that please ? ... And while I'm at it ;) What about negative prompts ? That's super useful too...
Cool~ but, how to use long prompt and custom model in invokeAI? thanks
i think a photoshop plugin based on automatic111 has more prospects
Where does InvokeAI look for its models? I've just got a blank drop down box.
Upvote for dark mode
When I click the license link for those two, I can't find the acceptance and agreement portion. May you teach me what possibly happened? Is there anything I do it wrong?
There's no license agreements at the two links - do they no longer require you to agree to them?