0. Control net. (god-tier) 1. Dynamic prompts 2. Can never have too many 'quality' meta prompts: (masterpiece), (extreme detail), etc 3. Always use 512 or 768 res and upscale with highres fix. 4. Don't get greedy; only about 3-5 Lora's can be supported per prompt before the colors start to melt. 5. xyz-plot with prompt s/r to test keyword strength effectiveness. Works on Lora's too! 6. start with a small prompt, then reuse the best seed and add more words piece-by-piece.
I don't think that would really work. the seeds would vary too wildly imo. Just run 20 in landscape, then switch to portrait and tailor the prompt and run 20 more. @@user-lk2tb9hx8c
Restore face is a great option for some applications, but it tends to remove things like freckles which may be desired. If this is the case, try adding "detailed face" to the prompt and turn off Restore Face. I've also found that "highly detailed face" produces results that look more like an illustration than like a photo. You can mitigate this somewhat by putting "photo of" before the positive prompt and adding "painting, drawing, illustration, sketch, anime" to your negative prompts.
I've just gotten started with the whole stable diffusion stuff and I'm enjoying the journey through the rabbit hole, thank you for this video it helped a lot!
Thank you for your tutorials they real are a great help. You have a very good delivery and pace, which makes understanding SD so much easer. Once again thank you.
Thank you very much. This video was very helpful in all aspects, especially the one I was looking for, the quick settings list. Like someone said before I appreciate your kind of style to present the videos. Thank you a lot.
Brilliant content for a beginner like me. There are SO MANY different settings with unknown meanings that it is hard to get things set up quickly. Thanks a lot. 🙂
CLIP_stop_at_last_layers Add that to your Quicksettings list like shown in tip 5. This will let you quickly jump clip layers. Very important. Many models/loras do not mix on same clip layer.
Hi @OneGiantLeapforAI, thank you very much for share with us this amazing tips. I would like to as you one little thing, I observe in my researches that the versions of SD has different performs and produce different results depending on with Colab you are using also if the version is using Gradio interface or not, and this with all version 1.4, 1.5. 2.0, 2.1, so I would like to ask you if you can say with version you are using in this video and where I can find the extensions. Best regards
First off, thank you!! Love these tips! But on that note, I've added some of these extensions and had my first issue with Automatic1111 running out of VRAM. Would that have anything to do with the new extensions or anything?
can anybody tell me why i can’t see any installed extensions on my webui, yes i can see them under my extension folders in directory but why there are not on my webui, i can’t use..
It usually looks for a vae with the same name as the model. Like x.chpt would woukd have x.vae.chpt. if not found in the model directory, then uses the default vae model. When no vae needed for a specific model make sure to use ''none''.
0. Control net. (god-tier)
1. Dynamic prompts
2. Can never have too many 'quality' meta prompts: (masterpiece), (extreme detail), etc
3. Always use 512 or 768 res and upscale with highres fix.
4. Don't get greedy; only about 3-5 Lora's can be supported per prompt before the colors start to melt.
5. xyz-plot with prompt s/r to test keyword strength effectiveness. Works on Lora's too!
6. start with a small prompt, then reuse the best seed and add more words piece-by-piece.
legend
I don't think that would really work. the seeds would vary too wildly imo. Just run 20 in landscape, then switch to portrait and tailor the prompt and run 20 more. @@user-lk2tb9hx8c
Love your style of delivery - no nonsense, not talking endlessly without demonstrating and no stupid gimmicks. Fantastic!
ive never had my mind blown so fast in my life
You pretty much nailed my "to-go-to" extensions! Also took your time to explain what they are. Thanks for a good video.
I've spent many hours days and weeks working in stable diffusion/a1111/deforum and 3 of these tips were brand new to me! Tysm!
Restore face is a great option for some applications, but it tends to remove things like freckles which may be desired. If this is the case, try adding "detailed face" to the prompt and turn off Restore Face.
I've also found that "highly detailed face" produces results that look more like an illustration than like a photo. You can mitigate this somewhat by putting "photo of" before the positive prompt and adding "painting, drawing, illustration, sketch, anime" to your negative prompts.
The aspectio ratio one is super simple yet extremely helpful!
I've just gotten started with the whole stable diffusion stuff and I'm enjoying the journey through the rabbit hole, thank you for this video it helped a lot!
Thank you for your tutorials they real are a great help. You have a very good delivery and pace, which makes understanding SD so much easer. Once again thank you.
Thank you very much. This video was very helpful in all aspects, especially the one I was looking for, the quick settings list.
Like someone said before I appreciate your kind of style to present the videos.
Thank you a lot.
Brilliant content for a beginner like me. There are SO MANY different settings with unknown meanings that it is hard to get things set up quickly. Thanks a lot. 🙂
Amazing tips! Thanks for sharing. Now if only I could figure out this stupid CUDA out of memory issue...
really cool, thanks. more please
CLIP_stop_at_last_layers
Add that to your Quicksettings list like shown in tip 5. This will let you quickly jump clip layers. Very important. Many models/loras do not mix on same clip layer.
Thank you! great content!
great tips, thank you
1:35 i use the ratio selector. now I know there other option. ty
and Image browsing is neat and the website is great. ty again.
have a nice day
Dude, agian, very helpful tips. Thank you!!!
Sick tips and great extensions. I've suffered long enough calculating my aspect ratios 😂🎉❤
Great tips and a real time saver
Straight to the point. Nice!
Thank you, and thank you commenters for extra tips 👍
I appreciate your content.Thanks
Thank you so much for these tips, they are all very helpful!
Nice additions. Thank you.
Thank you - didn't know about any of these...
Wonderful content and channel 👏👍
you deserve millions of subscribers
Love it. Please keep it up. Really useful tips.
Very nice, sober and useful thanks a lot
Dude, where have you been all my life?
Great stuff man.
After launching SD, press the arrow button under "generate" and it fills the prompts and settings with the latest generation...
my Stable Diffusion Extensions tab doesn't have a search bar :( .... And aspect ratio helper is not listed for me
Hi @OneGiantLeapforAI, thank you very much for share with us this amazing tips. I would like to as you one little thing, I observe in my researches that the versions of SD has different performs and produce different results depending on with Colab you are using also if the version is using Gradio interface or not, and this with all version 1.4, 1.5. 2.0, 2.1, so I would like to ask you if you can say with version you are using in this video and where I can find the extensions. Best regards
awesome, thanks. Y|ou say 5 more but I can't find the video with the first ones? Any help?
First off, thank you!! Love these tips! But on that note, I've added some of these extensions and had my first issue with Automatic1111 running out of VRAM. Would that have anything to do with the new extensions or anything?
Super! BIG FANX!
🤩
is there a save settings add-on so i don't have to change and set everything each time i load a1111
can anybody tell me why i can’t see any installed extensions on my webui, yes i can see them under my extension folders in directory but why there are not on my webui, i can’t use..
I receive "PermissionError: [WinError 5] access denied" when I try to install the aspect ration helper, and yes I'm running webui as admin
i wish the words-to-use site have definitions for all the provided words.. i have a hard time checking what some of the words meant
The right-click thing never worked for me. I just see the standard right-click menu that appears for every other website.
This video is only a month old and I can't see any such quick settings option that you show here.
no such aspect ratio extension available
Thx
Does automatic VAE do what I think it does? I mean does it somehow know what the right VAE is for the model you're using??
It usually looks for a vae with the same name as the model. Like x.chpt would woukd have x.vae.chpt. if not found in the model directory, then uses the default vae model. When no vae needed for a specific model make sure to use ''none''.
how did you get that VAE dropdown menu at the top.. been looking for that one! thanks
He shows you how at 5:25
"yeah thats been here this whole time" WHAT
How to download vae
Right clicking absolutley mandella effect in action.... right? Xd
this "forever" is a old extantionbut......as always ALL DIFERENT....
omg I've been wasting so much time with aspect ratio calcuations.
The like was for the right click thing - could have made this a SHORT brother!
No, then I wouldn't have seen it since I ignore shorts. This was valuable as a video.
@@lpnp9477 Oh I am a fan and a subscriber. I just meant that one trick could change lives! And fill up hard disks!
no way... right click. generate forever! ahhhhhhhhh