What I WISH I knew when I started (Stable Diffusion)
Вставка
- Опубліковано 12 бер 2023
- Timestamps below:
0:21 Start a "keep for prompts" folder. Many PNG files store EXIF metadata with the prompt and settings for that image. Use the PNG info tab to view it, and send to txt2img to img2img.
1:39 Read the description for your model! The creators know how to prompt.
2:25 Start a styles library early! Use the model name with the style. Also HOW TO DELETE STYLES at 3:47
4:32 Use DDIM for early iteration. You get good results with as few as 10 steps, so this is great for people with slower GPUs, or if you are early on in the prompt iteration process.
5:40 Use ctrl+arrow up and down to quickly change (emphasis) in your prompt.
You really looked all the way down here? Well, here's your reward: my Discord server that it totally not ready for release yet. It's literally empty but you can join it if you want. / discord
There's a real opportunity to grow a UA-cam channel with this content... Good luck to you!
It’s really encouraging to see comments like this. Thank you!
Thanks for the tips! Another great hidden gem of A1111 is if you right click the generate button, you can select "Generate Forever" and then right click again to cancel it. That way you can leave it overnight or whatever and create a ton of variations without having to mess with the batch counts. Also, if you change any parameters (including the model) during generation, it uses those settings on the next image. Cheers!
🍻
Thnx, last one is very useful. You also can drop PNG to prompts arean and then click arrow above style
Nice! So many little details to learn.
This is simply one of the best "knowledge distribution per minute example"!! Thanks much for sharing!!!
Useful, some I'd worked out however the styles is useful for default / negative prompts that I always end up re typing... that weight tip ctrl+arrow :). thanks..... My own tip...on the page where you click to insert a lora and it inserts the lora prompt. If you hover the mouse over the lora you want there is a small i icon in the top corner, click the i and it shows the key words and a weight number next to the key word that triggers the lora, some lora have only 1 trigger word just to make it work, others have a big list that add many different effects.
I was today year's old when I finally learned these tips. I wasn't far into this video at all when I was absolutely blown away. You have no idea how valuable this information is to me. How life changing this is. This is going to help me so much. This is literally so easy to understand, and for new users and even seasoned users of this software, it's absolutely required viewing. I can't thank you enough for it.
Life changing?
When you come back to the computer next day and start with a blank prompt. Hit the blue left most icon under Generate to load up the settings from the last created image, so you can continue from where you last left off.
I think it's pretty important to understand that CFG scale is connected to sampling steps. The higher the CFG count, the higher you need to set the sampling steps to not get strange dark images. So knowing that, you can set the CFG scale to ~3 and set the sampling steps to something like 10 to get good results. But they will be lighter than with higher CFG scales. But it's also good for fast iterations. With a 3090 i get an 512x1024 image in like 3 seconds
Didn't know about the control up and down arrow, great tip. Also, u can open up the styles csv file and edit it with excel instead of notepad. I deleted rows this way and copied/pasted new versions too.u can reorder your styles easier in excel too.
These were some great tips man. Thanks a lot :) You have a good way of explaining things and in a calm manner. No rapid quick cuts and odd editing. You get right to the meat of it as well. Much appreciated and you got a new subscriber.
I found this video and definitely a unique take on saving the time and really effective shortcut that is useful to keep. Thanks, keep on sharing small golden nuggets like this
Actual lifesaver! So much stuff I wanted to be a feature but had no idea they existed lmao. Thanks legend
Super helpful! I like the one about reading the manual on models I pick up for workflow tips!
Thank you sir! The retreiving of the PNG Info is a great feature. I've been cutting/pasting my prompts from a text editor to try and recreate images I saved. The save styles is great too.
Great tip, the last one. Thank you!
Great and (extremely valuable) upload for new starters and folk still trying to create amazing content. 👌👌👌. Thanks
I was just looking at when you last did a tutorial it was 5 months age. I hope all is well with you and you'll take One Giant Leap back to us. Your videos are invaluable to us.
The DDIM tip is mindblowing. At list for me. I don't have a GPU card, consequently my generations take too long. This will help me save a lot of time.
The weight tip was great too! Keep the good work!
Please continue like that, thank a lot for sharing.
pnginfo ! you're a god amongst youtubers
Concise and really useful tips, thanks!
excellent tips! Last one was a gem, wow
Great video. Keep it up. I like your style.
So good. Loved it. Would love to know more about prompts, prompt variations and prompt tricks. Thanks
WOOOOOW. Many MANY, thanks man 🚀
Thanks for making this content. I was already saving prompts on a word doc 😅
wow that was incredible thank you man
Great video!! Thanks!
Saved the best for last. The DDIM to speed up ideation/rough work is excellent. The last tip is invaluable. I will use that allot now... =]
Great stuff!
Great advices, thanks mate
This is so helpful. Thx bro
Thanks, these are very useful indeed.
awesome video helped me a lot
The last tip made my tip release fluid
I’m here to please.
so helpful thank you!!!!
the last 2 is new to me, thanks
thanks for the indeed useful tips 😊
Good tips!
thank you so much 👍
Thuis was extremely useful info, thanks
Thanks ! 👍
+100 for the Ctrl shortcut tip
perfect !
cool - thxx!
Just found your channel! Great pool of creativity and knowledge here! Thank you! - One question: Why "One Giant Leap"?
Great collection of tips. Today's year old for most of those.
Another thing I do is "Pin to Quick Access" the folders that I frequent in Windows
yo that png info one i didn't know. i had an insane artwork but i lost the prompt so ty.
The last tip was good
You are an angel lol
That’s so weird. I just made a folder called keepers last night and it’s your first tip. 😮😂
👀 Protogen - Semirealistic ^^
There's a FOLDER OF ALL YOUR IMAGES!!?!???!?!? OMG!!!
Nice, should do more
Believe me, they’re coming! What would you like to see?
when i put in a png to get the promt, it says that there are no perameters. Is there something i'm doing wrong?
does png info work after an image was processed thru something, edited or upscaled by an outside program?
It's called "EXIF data" and many programs will preserve the data, but without specific information on the outside program you can't know for sure.
my attemps are looking really ugly , I have try so many thing but something as simple as seamless pattern of cat, look like the cat felt in acid. Lol. Help!!! can it the most important thing to change to make this kind of generated style??? any ideas???
dpm 2 karras nvr works for me,
where can I download stable diffusion?
Pro tip:
I was trying to recreate lackadaisy style…
Do NOT use the prompt “((((((((((((((((anthropomorphic))))))))))))))))))”….. bad idea
What’s the max strength you can give a prompt? I’m not sure if there is a limit to it. I always just go for (x:1.5) but it could potentially be higher. Also what does doubling or tripling the parenthesis do, and is there a limit to how much you can use that as well?
As far as I know there is technically no maximum strength, but above (x:1.4) will negatively effect your outputs. No reason to go above 1.4 unless it's with a LORA, which can use a larger positive and negative range. Number of parenthesis is the decimal digit, so a ((((four)))) parenthesis emphasis is the same as (four:1.4)
@@OneGiantLeapforAI How you mean with Lora ? Lora embeddings can go only in range 0-1 , how strongly Lora weight will be use with prompt, going over does nothing. Also if you use multiple they always need to equal to 1 , so if you have 2 lora embeddings you gotta set it 0.5 & 0.5 as example.
500th like
Perfect video, thank my friend!
Soydev norwood