Hello everyone, Keyboard Alchemist here! This is episode2 of my stable diffusion/automatic1111 tutorial series, where we go through a simple workflow designed to help understand the Sampling Steps, CFG scale, and Sampling Method parameters using the XYZ plot tool, and get the most out of that new Model you have just downloaded! I hope you enjoy this video. Feel free to leave a comment below.
Oh man... I hope to see this tutorial sooner. I was doing all that manualy. But hey, at least design is better xD Thanks for your work! BTW... is there any ways to add time of generation into it?
I'm glad this was helpful for you! I don't know if there is a way to list out time of image gen in the XYZ plot. That hasn't been an use case for me. I usually just look at the command line progress print out.
Great content! Haven't seen many other vids that are as clear as this one when explaining some of the underlying concepts for image generation using various samplers. Definitely will be taking advantage of the x/y/z plots moving forward!
i dont get it, on the steps in xyz, you can have it 10,20,30,40. but what about the one on top, if i change the top to 10 or 20 or 30, it would make different images as well, how would i know which to use without the xyz? same on lora, my lora weight on top is 1, on bot is 0.2,0.3,0.4 etc, if i have the top to 0.9 or 0.8 it would make different images compared to my wirhgt lora of 1 on top
When I started I just wanted to make beautiful things as fast as possible. Now that gotten a little out that our of the system, I really wanted to learn more about what things in the various UI's mean and how to use them efficiently. I'm still learning a bunch everyday, but man.. this helped a ton with workflow decisions. Several other creators have suggested ranges very much in lock step here. But, here, you helped demonstrate the underlying decision trees. Thank you for the deep dive here.
Thus far, I have never had a single model NOT work great with: cfg 6.0, 30 steps, dpm++ 2m karras. No, I tell a lie - there are some models that simply don't work at all, but that's the models' fault, not the sampler/step count/cfg.
I stumbled upon one of your videos searching for upscalers, after going down the open source webui forge road. TH-cam is smart like that. Frankly, I was somewhat confused and quite bored, though I am also tired, but what did shine through was that you are doing a deep dive into the whys and not just the hows, which empowers users... teach a man to fish, sort of thing. Keep it up. I foresee good things for this channel... and trust me, there are tons of up and coming stable diffusion channels popping up left and right every day, and still I say that, because it's the same content mentality that I apply, and a standard I set for myself: leave no stone unturned. Once every possible question has been answered, which requires testing since no one's actually bothered to do that, then it's good enough. And taking that extra step, well, it shows.
Very useful! I wonder if there's a similar way to automate testing for a model's optimal parameters for animation. Like, a plot of the 30th img2img iteration with various denoise & cfg values.
I've not done much with animations in Stable Diffusion, but your idea sounds interesting. Please share any insights that you may get from testing it out. Thank you for watching!
@@bunnystrasse Ha! It's a math degree. One course I took, called design of experiments, looked a lot like this x, y, z method to hone in on an optimal design.
![Tiled Diffusion with Tiled VAE / Multidiffusion Upscaler, the Ultimate Image Upscaling Guide [A1111]](http://i.ytimg.com/vi/44waH3sDYOM/mqdefault.jpg)
![Tiled Diffusion with Tiled VAE / Multidiffusion Upscaler, the Ultimate Image Upscaling Guide [A1111]](/img/tr.png)
Hello everyone, Keyboard Alchemist here! This is episode2 of my stable diffusion/automatic1111 tutorial series, where we go through a simple workflow designed to help understand the Sampling Steps, CFG scale, and Sampling Method parameters using the XYZ plot tool, and get the most out of that new Model you have just downloaded! I hope you enjoy this video. Feel free to leave a comment below.
Great video.
It‘s absolutely amazing!
Thank you!
useful video thanks for this ! 👍
I'm glad you liked it. Stay tuned for more tutorials!
Oh man... I hope to see this tutorial sooner. I was doing all that manualy. But hey, at least design is better xD
Thanks for your work!
BTW... is there any ways to add time of generation into it?
I'm glad this was helpful for you!
I don't know if there is a way to list out time of image gen in the XYZ plot. That hasn't been an use case for me. I usually just look at the command line progress print out.
Great content! Haven't seen many other vids that are as clear as this one when explaining some of the underlying concepts for image generation using various samplers. Definitely will be taking advantage of the x/y/z plots moving forward!
Thank you for your kind words! Stay tuned for more quality content!
Agreed! Really2 loved this vid content, it explain and helps a lot!! Keep it up!
@@eggyryan5610 Thank you very much!
i dont get it, on the steps in xyz, you can have it 10,20,30,40. but what about the one on top, if i change the top to 10 or 20 or 30, it would make different images as well, how would i know which to use without the xyz? same on lora, my lora weight on top is 1, on bot is 0.2,0.3,0.4 etc, if i have the top to 0.9 or 0.8 it would make different images compared to my wirhgt lora of 1 on top
When I started I just wanted to make beautiful things as fast as possible. Now that gotten a little out that our of the system, I really wanted to learn more about what things in the various UI's mean and how to use them efficiently. I'm still learning a bunch everyday, but man.. this helped a ton with workflow decisions. Several other creators have suggested ranges very much in lock step here. But, here, you helped demonstrate the underlying decision trees. Thank you for the deep dive here.
I'm glad to hear that the video was helpful to you! Thank you for your kind words! Cheers!
Goood,!!
Thus far, I have never had a single model NOT work great with: cfg 6.0, 30 steps, dpm++ 2m karras.
No, I tell a lie - there are some models that simply don't work at all, but that's the models' fault, not the sampler/step count/cfg.
Very true, I like the DPM++ 2M Karras sampler for this reason.
CFG = Classifier Free Guidance? I have been calling it "config" all the time. People should stop using acronyms
THis is insanely helpful
I'm glad it was helpful. Thank you for watching!
God, this was so well done.
Straightforward, and SUPER clear on everything.
Sub'd
Thank you for the sub! I'm glad you liked the video.
HI how do you generate the heatmap?
Thanks for this! I'll buy you a coffee!
Thank you for your kind words!
I stumbled upon one of your videos searching for upscalers, after going down the open source webui forge road. TH-cam is smart like that. Frankly, I was somewhat confused and quite bored, though I am also tired, but what did shine through was that you are doing a deep dive into the whys and not just the hows, which empowers users... teach a man to fish, sort of thing.
Keep it up. I foresee good things for this channel... and trust me, there are tons of up and coming stable diffusion channels popping up left and right every day, and still I say that, because it's the same content mentality that I apply, and a standard I set for myself: leave no stone unturned. Once every possible question has been answered, which requires testing since no one's actually bothered to do that, then it's good enough. And taking that extra step, well, it shows.
Thank you, I appreciate your kind words and encouragement!
You are awesome! This helped soo much. Thank you for the great content!
I'm glad this video helped you! Thanks for watching and stay tuned for more quality content!
Can XYZ parameters be saved for future recall?
Unfortunately, I don’t think that’s possible right now. But that’s a good idea.
Thank you! I just dwelve into SD and I can't stop exploring what it can do (mostly arch viz).😲
That's awesome! Stable Diffusion is really great! So much community create extensions that works like magic. Stay tuned for more fresh videos!
Very educational. This was great!
Thank you very much!
Very useful! I wonder if there's a similar way to automate testing for a model's optimal parameters for animation. Like, a plot of the 30th img2img iteration with various denoise & cfg values.
I've not done much with animations in Stable Diffusion, but your idea sounds interesting. Please share any insights that you may get from testing it out. Thank you for watching!
This is great. You must have studied ops research. Looking forward to more!
I'm glad you liked this video. Thanks for your support! I will have more videos coming soon.
What’s ops research
@@bunnystrasse Ha! It's a math degree. One course I took, called design of experiments, looked a lot like this x, y, z method to hone in on an optimal design.
man! what a great video ! so glad i've found it. great work
I'm glad you enjoyed the video! Thanks for your support!
Clear and concise, impressive lesson.. Thanks !
I'm glad you enjoyed the video! Thanks for watching!
Tres bon tuto, je m'abonne
Thank you for your support!
Great high quality content!
Thank you for your kind feedback!
This has got me curious to try CFG 30 steps 150, looks like it has a lot of extra details in it
Happy experimenting!
What did you find when you tested this?
@@q5sys Didn't like the results, but I did very little testing to be honest
@@androsforever500 Thanks for the response, I appreciate it. I might have to try some testing myself if I can find the time.
Thank for the lesson.🫡🫡🫡
Subscribe!!!
Thank you for the sub! Stay tuned for more tutorials!