InvokeAI - Fundamentals - Creating with AI
Вставка
- Опубліковано 5 жов 2024
- ⚠️ Note: This video features a previous version of Invoke's UI.
We’ve recently updated our interface, and while the core concepts in this video still apply, the visuals may differ from the current version. For content featuring the latest UI, please visit our channel and check out our recent videos.
InvokeAI - Fundamentals - Creating with AI
github.com/inv...
For those just getting started with Invoke, or for users looking to level up their understanding of how the technology works in advance of the full 3.1 Workflow release, the Invoke team is excited to cover the "Fundamentals" - With more coming soon.
Note that this includes some heavily simplified analogies, especially around models. It's a great way to get started, but there is some technical nuance that we didn't cover in this video.
Make sure to download the latest installer from our Github repo.
github.com/inv...
Join the Discord
/ discord
Thanks from the InvokeAI Team
It wasn't too technical and I was shocked and sad that the video ended. It was well explained and we need more.
I do not know why you guys are working so hard, to create a product that is free, and then working just as hard teach us how to use it... but thank you so much. This video was VERY helpful! Liked and Subscribed.
I've certainly heard starting step and ending step explained before but I wasn't confident enough to do little more than default generations. Because you set up the foundation beforehand I can honestly say I have a firm grasp on how the program works now. Using two of the same control net models with one focused at the beginning steps and the other toward the end kind of blew my mind. Thanks for putting this together.
This has been the best video on explaining these tools.
You can even get more technical, it was very well explained!
wow - incredibly helpful high level explanation of the diffusion model process. i have often struggled to understand denoising and the the percentage values of controlnet working in comfyui. will definitely check out more of your vids for sure!
It would be great to see a continuation of this; Showing creative ways to use two control images of different types; showing examples of the other options and effects it has on your images (ex. Balance vs Control vs Prompt), etc...
This was insanely helpful . As a teacher myself, I think you did a wonderful job of pacing, visual aids, and explanation. This was so helpful for me to learn what I can do and how to get better results! 😀🤩
This was a perfect not overly technical explanation. I started playing with SD fairly recent and haven't yet got to know all the available tools too well, this video helped me fill in some gaps and i would ABSOLUTELY like more videos in this style about different SD topics
the video was just right for me in terms of complexity. thank you very much!
Great video. It's hard to find good content that describes conceptually what all the components are in the ecosystem of Stable Diffusion and how to use them. Please keep these kinds of videos going! Don't be afraid to get even more technical, but this was still very helpful.
This is absolute gold content! Thank you for this 10/10 video. Please make more of them. There are lots of settings, workflows, features and terms i want to understand. Thanks so much
Nice! Been away from Invoke for several months as A1111 was simply so much more advanced. Great to see that you guys have rolled out a version with Control Net. Might be time to switch back to Invoke.
Thanks for a great explanation. You're one of the best teachers, whether it's Invoke AI or any related subject to AI art, I always come away with a much better understanding of the whole process.
Really impressed with InvokeAI's work showcased in this video! I genuinely enjoy the content you put out and the way you break down technical details. I hope to see more in-depth technical videos from you in the future. Keep up the excellent work!
Just a casual and occasional user, and this really helped clarify concepts differently than the head-cannon I had created. Please keep these coming, as they are in that junction of useful without being too technically advanced.
Please, keep pushing videos like this one. It's perfect as it is!
Excellent video… It was detailed, informative and well structured
I just love InvokeAI and really got addicted to it. Thanks a lot for everything guys, this is awesome! 28s on a M2 Airbook for an image is not too bad. Kind of jealous seeing how the workflow changes when you have a really good GPU in the other Videos. Top Notch explanation!
man you are a life saver, that video is too damn good to be free
Really well explained! While I played around with Lora's, TI's etc I never used controlnet since it seemed so abstract but this gives me a good idea of how the logic works behind it, really usetull!
Please make many more of these. Lovely video. I’m really hooked on this product.
This was really helpful! Im just turning knobs at random and hope for the best. This gave a lot of insight. More please!
Best Diffusion explanation I've Ever saw! TNX
Great video my guy. Learning invoke ai right now and it is amazing.
Amazing video! I'm glad to see more videos with technical stuff
Loved this, super informatively structured.
This was quite easy to understand. Very helpful.
AMAZING! more of these please
More technical fundamentals please! :P :D
HOLY CATS! That was a TERRIFIC explanation!!! Was sad to see it end so soon. I was left wanting MOR… Hey waitaminute… I see what you did there. ;P Can’t wait for the sequel. Things are much clearer now. Thanks for this. Really appreciated the dictionary analogy. Something clicked for me there.
This was explained plainly and clearly just how I like. I would of liked too learn how to exactly use diffusers via hugging face... Upgraded the GPU to gimmie access to the powerful and fun stuff
Amazing tutorial! Thank you so much for taking the time to release these videos :)
That was incredibly informative and interesting.
The best explanation I found so far. Thanks!
Just great !!! What a work you did !!! GG
Very helpful thank you.
GREAT video! It really helps to have a better understanding of this diffusion process. Could you provide similar information for Lora's, embeddings, ... in a similar style?
a specific tutorial about controlnet would be cool , what is the different between the processor option ( depth (midas ) depth Zoe and all the other options ?
i just learned that when i use simple picture of a 3d shapes and use the Processor ( Depth Midas ) i get way better results . coming from 3d i just thought "hmm this makes more sense because invoke gets more info about the real shape " i still dont know if im thinking the right way .
The ControlNET models were trained to understand how certain pre-processed images map to the structure/output of the resulting image. Many of the processors can be mixed/matched and used with different models, even if they weren't trained on them, if the type of the image is very similar in nature (e.g., edges, depth, etc.)
Great video, keep em coming!
more technical !!!!!!!!!!!!!!!!!!!!!
very well explained ❤
it was amazing, I learned a lot
Fantastic video!
it would be super cool to have some mouse over info to see the different settings between the generated pictures in the beard gallery .
great explanation
Please an advanced NODES tutorial!! xD Thanks.
I love u 4 helping better understand control nets lol!!!
great Product... while everyone is talking about features... usability is the most important one, and here invoke roules
Never the less... my attemp to upgrade Invoke to 3.02.. left me with a broken installation, and i cant get it running again.
but still.. i hope this programm will receive more attention. because its important that also the non- nerds get their hands on this impressive technique
It would be so much easier if we could write all these things up in separate text boxes for each category or if we could categorize them ourselves clearly instead of giving vague descriptions.
I'm completely new to AI image generation and I find it fascinating so you can probably do what I just said already but can I clarify to invoke what the subject or subjects of the image are, example [Subjects: ninja, humanoid alien, robot" then add descriptions for each ninja: basically tell invoke what its subjects are, what they look like and then add quality modifiers and aesthetics under their own field. Because adding complete descriptions with words that just come to you doesn't seem very productive in the long run.
Great vid
a nodes editor tutorial would be super cool
Apologies if this was already answered but when doing image2image what's the difference between the original uploaded image and a seed image?
Does the selected seed image replace the original image or is the generation process still utilizing/drawing from the original image when invoking?
(I'm very new to all of this)
Thanks!
Is there a setting somewhere in InvokeAI equivalent to AUTOMATIC1111's Clip Skip ?
this was great
is it possible to have the models as downloadable chunks (zipped ofcourse) or better yet a torrent
the reason im saying this is, in cases where there is a cap on download speed and how long you are connected to the internet, it become pretty useful to download in chunks even over days
please look into this
kind regards
i feel like you should assume everyone watching your videos it may the first time ever hearing about InvokeAi and therefore should introduce what Invoke Ai is each time before diving into a tutorial. I watched multiple videos trying to understand what this is and couldn't so even though i wanted ai videos i just gave up.
This is madness.
(2:12) How does it understand the difference between "weed" and "weed"? Also, what if the guy writing your dictionary doesn't know the difference between a weed and a flower? Also, also, what happens if there are 10,706 different types of "weed" in the world? Also, also, also, why the hell would you waste time and space filling your dataset with "purple" when you can just define all the hex values in an enum for your parser?
(2:13) "off-center" has a hyphen in it.
(3:18) Confessing that you were too lazy to create a synonym parser isn't upselling your product. How can I recommend your product because it gave me a great "running" image if the great running image was filed under "sprinting" but I didn't know that?
(3:23) "Candycane" is two words. "Award-winning" has a hyphen. How does the "dictionary" know what "award-winning" means? That's not an objective evaluation. I saw American Beauty. That movie was crap. Still "award-winning" though.
(3:34) Why are there "quality" modifiers? Why aren't you giving us the best quality by default? Why do I have to ask you for your best work? "Sketch" and "pencil art" are "styles".
(4:10) How could your robot POSSIBLY know what "don't make it mysterious" means? That's absurd. What if there was critical structural data contained in the shadows of the "mysterious" version of "gingerbread house"?
(6:48) Why are you splitting "sunflower" into "sun" and "flower". That's not what they asked for. That wasted bandwidth and CPU time from users begging and pleading with your system to understand the English language is probably costing you a fortune.
(7:50) You'd save a ton of CPU cycles and electricity by asking users for approval of a "10% denoised" preview image so they can interrupt the process if it's heading in the wrong direction.
(?:??) Why---on---Earth---are there no delimiters? Can't you see how breaking apart ideas with semicolons could change everything and give you a massive advantage over the market?
Wow this is a lot
Frankly I actually don't think it understands that, a weed is just something undesired by humans it not a biological term, a weed can still be a flower.
Will respond, assuming that your questions are in good faith and that you're interested in learning more!
- The rough answer here is "context". The denoising process looks at words in relation to the other words in the prompt. However, this is imperfect - If you prompted "a financial vehicle", especially with older models like SD1.5, you are just as likely to get a car in the image as something resembling what you are prompting for. The training process is identifying patterns, but having clearly distinguished labels for certain concepts, as you've identified, matters.
- Thank you. Recognize that while I won't go back and re-record this video, I will submit myself to seven lashes for this error.
- A "synonym parser" is not really what is implied here - Moreso, taking a simple prompt and expanding on it using prompt terms that have been positively received by other users or in an art style that it believes you want, but not telling you what those terms are.
- It does not understand what these words really mean, only the patterns it has observed. So, for example, it knows that "-punk" as a suffix to a word, has a noticeable pattern (observed over "cyberpunk", "steampunk", etc.) and can generalize that to new combinations like "forestpunk" and "wizardpunk". Some of these will be comprehensible and useful to us, some won't. This is why understanding how this stuff works allows you to be more creative! Also, Christian Bale cried a little bit at your comment.
- Quality modifiers are a term we've coined to describe terms that generally result in positive results - However, the qualities of the image the user is looking for varies and are contextual in nature. This also goes back to "adding prompt terms to the users prompt to get higher quality without them knowing" - This is what others do, and an intentional decision we've made in order to provide more control. That said, we do have some ideas of how we can help users identify and use these more easily.
- "Mysterious" is an image label/tag that often has a very common aesthetic quality to it - Images that are darker, include fog, have a dramatic sense to them, etc. In this context, we're saying "none of that, please". Just recognize, it's not a robot - Under the hood, all machine learning is really just very fast/complicated math.
- I think you misunderstand how the technology works, so there is a lot that can be addressed in this one comment. The one I'll touch on - This software is free and open-source, and the majority of people using it are running locally on their machines. Our hosted offering at invoke.ai is primarily for professionals, and those users (especially enterprises) are taught how to take advantage of this technology to re-train models, understand the tool, and leverage it in their workflows. In any case, this isn't magic, it's a tool - And while improvements on interpreting what users want to see is a constant research endeavor, we see our job as helping educate people on what is possible with the technology that exists today, and providing them the capabilities to integrate it into their creative pursuits. We also think that this technology is revolutionary on a societal scale, and that providing high quality open-source software to leverage it is the right thing to do.
- We preview the entirety of the denoising process live, as it's happening, and users can cancel it if they don't like the shape/composition/etc. I do that regularly! (Also, this technology uses GPUs more than CPUs)
You know, this comment itself might just be as educational as the whole video! Thanks for prompting us (ha!) to write it.
@@invokeai In the sentence "And while improvements on interpreting what users what to see" you misspelled "want". Alongside the fact that you didn't put a hyphen in "off-center", I really don't understand how you don't delete this channel and go back to sleep for 30 years.
Sarcasm aside, your reply to such a hasty comment is admirable, and I'm all here for it. Love the content
@@invokeai Thanks for responding. I appreciate the clarifications.
prompt idea: shit in the fan
is this subscription based or limited by payment cause Im not paying for monetized trash
it's free & opensource
bro this is free and the easiest way to install a stable diffusion generator on an average consumer PC
I recommend this kinda software on everyone's PC if they can get it