Any reason you are using LMS at 50 steps for your starting point? I generally run Euler A at 20 and then, once I want to dig deeper, DPM2 a Karras -- LMS so far hasn't generated great results, but I might be doing something wrong :)
I like the consistency of k_lms. Euler_a gives you a good result at 20, yes, but images might vastly differ between 19,20 and 21 steps :) k_lms at 50 is a great starting point for beginners. It does however struggle with cfg scale about 15.
@@sebastiankamph thanks for your reply! I have been looking for a stable sampler and so far DPM2 a Karras (is there a shorter name for this?) seems to "stay on target" more consistently than others - but will give LMS a try today and see how it performs
erghh... - those provided kind of crap results to be honest :/ and im a little bit surprised. Haven't used midjourney for a long time now but I guess it will do much better
These image generation ai's started off as accessible to most everyone - no coding was necessary, but one tended to need a familiarity with computers and such above that of a typical end user, to be able to navigate the inevitable hiccups one would encounter while making use of those early notebooks. Then the shared colabs started getting easier, more user friendly, and making the adjustable parameters more accessible to everyone. Not surprising, this was followed by a surge of new people getting into ai art, and expanding the community. I have mixed feelings about that tbh, but there's something else I've noticed which is, to me, a more significant bummer. From as early as I've seen, this community has carried on with a tone of open accessibility. All anyone had to do to make ai art, was discover where to find a notebook. There was really no gatekeeping, and not many exclusionary limitations to prevent anyone at all from taking part in this new and fun creative outlet. This is no longer the case, unfortunately. For both Disco and Stable Diffusion, they seem to me to be trending more and more towards exclusivity. And there came a point in time where limitations began to arise on both platforms. I'm not a professional artist. I do ai art for fun; purely for my own amusement, basically. So I've only ever made use of the tools available at no cost. I wasn't doing this very long however, before I noticed the limitations of being on the free-way. Those who were paying for upgraded colab subscriptions were able to put a lot more time into the practice of prompt crafting. And with access to faster gpu's through colab pro, their engagement was enriched even further. Having paid for a more efficient workflow, many began to pull away from the rest of us with regards to the quality of their images. Clearly, shelling out money for upgraded services was proving beneficial. Then when SD dropped, there were various voices putting an explicit emphasis on being accessible to everyone. In the beginning, it was. And to an extent, it still is. Just like before, however, certain limitations on that promised accessibility began to sprout up. Stable, lauded as being all-inclusive ai art for all, has ironically become the most restrictive and exclusive platform of all - even to exceed midjourney. Every advancement coming down the line nowadays can only be enjoyed if one has SD running locally. This requires a sufficiently powerful gpu, which is cost prohibitive for most people. Instantiating all the requirements to run SD on your own computer needs super user level knowledge, beyond that of most people. The ivory tower is only growing taller every day. And it seems we lowly peasants are shit outta luck. Just hafta learn to cope, I guess. AI art has taken a direction which is divergent from the ideal with which it began. That's kind of a bummer, I think..
I think you have some interesting points here, and yes, in some ways the AI art space have been pretty hard to access. But in my opinion, each new tech leap, while being only accessible to the top in the beginning, actually does get optimised for everyone in the end. Most of the time there's money involved to get it out to the masses, which isn't the greatest incentivizer. I think the SD community at large is for "accessible to everyone", but the community doesn't always pull the strings. By the way, have you checked out Stable Horde? They're crowdsourcing resources to let everyone enjoy SD freely.
@@sebastiankamph I think I'm just miffed because I'm not able to keep up with the pack. If it was the case that all the cool new functions could only be implemented if you could modify the code yourself, I don't think it would bother me as much. Because if I wanted to use the new functions, I could either learn to code, or don't - in which case _I_ am the only thing holding me back from doing so. That, to me, is an acceptable trade off. I can learn to code if I want to. And I don't think there are any forces beyond my control which would prevent me from learning, if that's what I wanted to do. But it's another matter entirely to be locked out by a paywall, basically. Understand that improving one's financial situation can't be done with a similar effort as learning to implement a few lines of code. I have not heard of Stable Horde, but I'll look into it. Thanks for bringing that, and a few other things to my attention. I appreciate the effort you put in to relaying helpful information to those seeking it out.. 👍
I was dying laughing when you said, " It looks like shit".
Hah, glad I could provide you with a moment of happiness 🌟😊
I love seeing these process videos so much!
Really happy you liked it! 👑
very cool and creative workflow
Thank you! Glad you liked it ☺
That was really helpful to see your process.
Glad you enjoyed it!
0:53 how do i see the image as it being made like that????
th-cam.com/video/_4rY0oPbUYA/w-d-xo.html
great..!! thanks!!
You're very welcome :)
if only it really generated that fast XD😝
Haha, soon!
Any reason you are using LMS at 50 steps for your starting point? I generally run Euler A at 20 and then, once I want to dig deeper, DPM2 a Karras -- LMS so far hasn't generated great results, but I might be doing something wrong :)
I like the consistency of k_lms. Euler_a gives you a good result at 20, yes, but images might vastly differ between 19,20 and 21 steps :) k_lms at 50 is a great starting point for beginners. It does however struggle with cfg scale about 15.
Asking the real question here. Also stumbled about it because my GPU is not so fast so i try to use a quicker sample.
@@sebastiankamph thanks for your reply! I have been looking for a stable sampler and so far DPM2 a Karras (is there a shorter name for this?) seems to "stay on target" more consistently than others - but will give LMS a try today and see how it performs
@@mehWatever euler_a at 20 and ddim at 20 or even 10 will provide fast results.
🤩🙌👏
erghh... - those provided kind of crap results to be honest :/ and im a little bit surprised. Haven't used midjourney for a long time now but I guess it will do much better
These image generation ai's started off as accessible to most everyone - no coding was necessary, but one tended to need a familiarity with computers and such above that of a typical end user, to be able to navigate the inevitable hiccups one would encounter while making use of those early notebooks. Then the shared colabs started getting easier, more user friendly, and making the adjustable parameters more accessible to everyone. Not surprising, this was followed by a surge of new people getting into ai art, and expanding the community. I have mixed feelings about that tbh, but there's something else I've noticed which is, to me, a more significant bummer.
From as early as I've seen, this community has carried on with a tone of open accessibility. All anyone had to do to make ai art, was discover where to find a notebook. There was really no gatekeeping, and not many exclusionary limitations to prevent anyone at all from taking part in this new and fun creative outlet. This is no longer the case, unfortunately. For both Disco and Stable Diffusion, they seem to me to be trending more and more towards exclusivity. And there came a point in time where limitations began to arise on both platforms.
I'm not a professional artist. I do ai art for fun; purely for my own amusement, basically. So I've only ever made use of the tools available at no cost. I wasn't doing this very long however, before I noticed the limitations of being on the free-way. Those who were paying for upgraded colab subscriptions were able to put a lot more time into the practice of prompt crafting. And with access to faster gpu's through colab pro, their engagement was enriched even further. Having paid for a more efficient workflow, many began to pull away from the rest of us with regards to the quality of their images. Clearly, shelling out money for upgraded services was proving beneficial.
Then when SD dropped, there were various voices putting an explicit emphasis on being accessible to everyone. In the beginning, it was. And to an extent, it still is. Just like before, however, certain limitations on that promised accessibility began to sprout up. Stable, lauded as being all-inclusive ai art for all, has ironically become the most restrictive and exclusive platform of all - even to exceed midjourney. Every advancement coming down the line nowadays can only be enjoyed if one has SD running locally. This requires a sufficiently powerful gpu, which is cost prohibitive for most people. Instantiating all the requirements to run SD on your own computer needs super user level knowledge, beyond that of most people. The ivory tower is only growing taller every day. And it seems we lowly peasants are shit outta luck. Just hafta learn to cope, I guess.
AI art has taken a direction which is divergent from the ideal with which it began. That's kind of a bummer, I think..
I think you have some interesting points here, and yes, in some ways the AI art space have been pretty hard to access. But in my opinion, each new tech leap, while being only accessible to the top in the beginning, actually does get optimised for everyone in the end. Most of the time there's money involved to get it out to the masses, which isn't the greatest incentivizer. I think the SD community at large is for "accessible to everyone", but the community doesn't always pull the strings. By the way, have you checked out Stable Horde? They're crowdsourcing resources to let everyone enjoy SD freely.
@@sebastiankamph I think I'm just miffed because I'm not able to keep up with the pack. If it was the case that all the cool new functions could only be implemented if you could modify the code yourself, I don't think it would bother me as much. Because if I wanted to use the new functions, I could either learn to code, or don't - in which case _I_ am the only thing holding me back from doing so. That, to me, is an acceptable trade off. I can learn to code if I want to. And I don't think there are any forces beyond my control which would prevent me from learning, if that's what I wanted to do. But it's another matter entirely to be locked out by a paywall, basically. Understand that improving one's financial situation can't be done with a similar effort as learning to implement a few lines of code.
I have not heard of Stable Horde, but I'll look into it. Thanks for bringing that, and a few other things to my attention. I appreciate the effort you put in to relaying helpful information to those seeking it out.. 👍