I'm assuming when you do that it just puts it into the queue. I usually prefer canceling the existing one just so I can move right on to the next one. But that's good to know.
@@AIchemywithXerophayze-jt1gg @dimamitlenko1830 in forge ctrl-enter cancels current generation (1x ctrl+enter cancels batch after finish current image, 2x ctrl+enter cancels it immediately - in theory, it doesn't always work actually). But also there is neat trick: right click on generate button and you can turn on "generate forever" option (same way to cancel it). I often use this to let it generate continuously while I play around with settings. You can also use agent scheduler extension and configure your mouse buttons (if yu have additional ones) for generate and enqueue. Enqueue with agent scheduler is awesome, especially with mouse shortcuts, as you can very quickly add to queue many images while changing sliders. Rarely use XYZ plot because of that.
Hi really good video about sdxl lightning..... is a comparison planned between the different models? Wood be interesting.... Btw I'm the creator of the model Mjölnir
Not stupid at all. I get that question all the time. It's an extension for automatic 1111 and it works with Forge as well. It's called aspect ratio helper. github.com/thomasasfk/sd-webui-aspect-ratio-helper.git
awesome video, kinda jumping back into stablediffusion after I found forge and its the only one that runs pretty normal to my expectations with my gpu 10-15 seconds per image. I noticed you said you had a 3070 ti, and your images were going pretty fast. I have a new laptop with the RTX 4060 which I thought would be fairly quick. auto1111 and comfy ui take about 3-5 mins per image. Are there any adjustments in the setting you made to have it render faster? Also how did you get the aspect ratio for your width and height sliders?
I'm pretty sure I said 3080 TI, but if I said 3070 I'm sorry. I actually have a 3080 TI 12 GB card. I do have a couple extra command line arguments, join my discord and let me know when you get there, and I'll see if I can get those over to you. See if that helps it makes a difference discord.com/invite/nbXqhy23
I have a 2060 with 6Gb of VRAM, and my image generation for 1024x1536 with Adetailer is around 90 seconds or less, for a 50 steps 25 HiRes. With 1.5 models. You shouldn't be at 3 to 5 minutes with a 4060.
You know I've found that the models with LCM baked in tend to run as fast or even a little faster. The problem is that I found that the LCM models detail level is lacking a bit compared to the lightning models..
@@AIchemywithXerophayze-jt1gg Can lightning models br combined with Control net and IP adapter in automatic 1111 yet? Also, which model is best for artistic purposes and not just photorealism?
I sit and laugh.. Everyone is so excited about these "Speedy" Models, but all I see in most yt videos regarding these models is subpar results. Apart from bragging rights to your other geek friends, Im still unclear whats the point of creating super fast images that are only 1/3 of the quality of what an "Experienced" prompter can create?
Interesting response. It depends on the settings. Typically with these lightning models, at least with the juggernaut version, setting the steps to 12, and the high-res fix steps to four, that's better results than the standard SDXL models. So I'm not sure where you're coming from on that.
@@AIchemywithXerophayze-jt1gg It just seems that a lot of development is going into speed instead of improving quality. Like Hands lol Even if you could generate images 3 seconds faster but the quality is going in the opposite direction, I just don't see the point. fingers crossed there's a team out there realizing the real breakthrough is going to be in getting the small details right like flawless hands instead of shaving seconds off of subpar images.
I do agree with you on that point that speed seems to be the Focus right now. Honestly I don't blame them. With how long it takes to generate images, video and other things. People are just excited to see something faster. I think the quality will come.
Been experimenting with these all DAY! And I finally went all in on your prompt gen too :D Great stuff Eric~!
Thank you and I appreciate that. You won't be disappointed.
Thanks for video! By the way, you can press ctrl + enter, during generation and it starts a new one without canceling previous
I'm assuming when you do that it just puts it into the queue. I usually prefer canceling the existing one just so I can move right on to the next one. But that's good to know.
@@AIchemywithXerophayze-jt1gg Yes, indeed, this function is from A1111, I completely lost sight of the fact that this is Forge
@@AIchemywithXerophayze-jt1gg @dimamitlenko1830 in forge ctrl-enter cancels current generation (1x ctrl+enter cancels batch after finish current image, 2x ctrl+enter cancels it immediately - in theory, it doesn't always work actually). But also there is neat trick: right click on generate button and you can turn on "generate forever" option (same way to cancel it). I often use this to let it generate continuously while I play around with settings. You can also use agent scheduler extension and configure your mouse buttons (if yu have additional ones) for generate and enqueue. Enqueue with agent scheduler is awesome, especially with mouse shortcuts, as you can very quickly add to queue many images while changing sliders. Rarely use XYZ plot because of that.
So I knew about the control enter at least to initiate a generation, I had no idea about the other ones though, that's freaking awesome. Thank you!
Hi really good video about sdxl lightning..... is a comparison planned between the different models? Wood be interesting.... Btw I'm the creator of the model Mjölnir
Actually, yes, I was planning on doing a comparison video. Good to meet you. I look forward to trying your model.
And I'm looking forward to see that video and curious about the result👍😉
Thanks for video! Can I ask you a stupid question? The aspect ratio button is some kind of extension for Forge?
Not stupid at all. I get that question all the time. It's an extension for automatic 1111 and it works with Forge as well. It's called aspect ratio helper.
github.com/thomasasfk/sd-webui-aspect-ratio-helper.git
awesome video, kinda jumping back into stablediffusion after I found forge and its the only one that runs pretty normal to my expectations with my gpu 10-15 seconds per image. I noticed you said you had a 3070 ti, and your images were going pretty fast. I have a new laptop with the RTX 4060 which I thought would be fairly quick. auto1111 and comfy ui take about 3-5 mins per image. Are there any adjustments in the setting you made to have it render faster? Also how did you get the aspect ratio for your width and height sliders?
I'm pretty sure I said 3080 TI, but if I said 3070 I'm sorry. I actually have a 3080 TI 12 GB card. I do have a couple extra command line arguments, join my discord and let me know when you get there, and I'll see if I can get those over to you. See if that helps it makes a difference
discord.com/invite/nbXqhy23
@@AIchemywithXerophayze-jt1gg ah your probably right i just heard 3 something TI.
I have a 2060 with 6Gb of VRAM, and my image generation for 1024x1536 with Adetailer is around 90 seconds or less, for a 50 steps 25 HiRes.
With 1.5 models.
You shouldn't be at 3 to 5 minutes with a 4060.
How does the speed compare against using base cdxl with LCM Lora?
You know I've found that the models with LCM baked in tend to run as fast or even a little faster. The problem is that I found that the LCM models detail level is lacking a bit compared to the lightning models..
@@AIchemywithXerophayze-jt1gg Can lightning models br combined with Control net and IP adapter in automatic 1111 yet? Also, which model is best for artistic purposes and not just photorealism?
I sit and laugh.. Everyone is so excited about these "Speedy" Models, but all I see in most yt videos regarding these models is subpar results. Apart from bragging rights to your other geek friends, Im still unclear whats the point of creating super fast images that are only 1/3 of the quality of what an "Experienced" prompter can create?
Interesting response. It depends on the settings. Typically with these lightning models, at least with the juggernaut version, setting the steps to 12, and the high-res fix steps to four, that's better results than the standard SDXL models. So I'm not sure where you're coming from on that.
@@AIchemywithXerophayze-jt1gg It just seems that a lot of development is going into speed instead of improving quality. Like Hands lol Even if you could generate images 3 seconds faster but the quality is going in the opposite direction, I just don't see the point.
fingers crossed there's a team out there realizing the real breakthrough is going to be in getting the small details right like flawless hands instead of shaving seconds off of subpar images.
I do agree with you on that point that speed seems to be the Focus right now. Honestly I don't blame them. With how long it takes to generate images, video and other things. People are just excited to see something faster. I think the quality will come.
@@AIchemywithXerophayze-jt1gg what I find annoying though is that t doesn't really listen to prompts...