discussion AI tools and prompts whats your favorite

[User_57059]

Hello togwther, I am looking for to understand the AI promts and technicial differences.
Wjat are your favorite AI prompts and tools?

Thanks for your feedback

 

[jackassik]

There are quite a few prompts I like. For example I like girls with smaller breasts but when you type "small breasts", they are still quite big. I found that using "petite" is quite good to achieve that.
Also, you can use "full body" if it's showing you only face or torso. Additionally "from behind, from side, from front" for the general position of the model. AI like to generate Asians so typing "European" helps, although I also like "Scandinavian". I also like "bokeh" to create those blurred lights in the background and "people in the background" or "surrounded by people" or "blurred background" to get some interesting backgrounds.
It's this what you meant?

 

[User_57059]

There are quite a few prompts I like. For example I like girls with smaller breasts but when you type "small breasts", they are still quite big. I found that using "petite" is quite good to achieve that.
Also, you can use "full body" if it's showing you only face or torso. Additionally "from behind, from side, from front" for the general position of the model. AI like to generate Asians so typing "European" helps, although I also like "Scandinavian". I also like "bokeh" to create those blurred lights in the background and "people in the background" or "surrounded by people" or "blurred background" to get some interesting backgrounds.
It's this what you meant?

Yes that is what I mean. Do you ever have tried to use same prompts on same AI?

 

[User_57059]

Yes that is what I mean. Do you ever have tried to use same prompts on same AI?

Do you got same pictures or different ?

 

[jackassik]

Do you got same pictures or different ?

Do you perhaps mean seed? You can run the same prompt multiple times and get a different image each time if you just change the noise seed.

 

[User_57059]

Ok seed means additional radom number for the AI. Like in the old SIM CITY SNES for the random landscape.
In my opinion when you use same seed and same prompt you should get same picture

 

[User_57059]

Ok seed means additional radom number for the AI. Like in the old SIM CITY SNES for the random landscape.
In my opinion when you use same seed and same prompt you should get same picture

Sorry jackassik for my stupid thinking around the corner

 

[TemperTempah34260923]

in my experience, the negative prompt field is just as important as the positive one.
you may have noticed there's also important modifiers , specifically the use of brackets and numbers next to a prompt.
LORA's and other add ons are extreamly helpful so i advise you to learn about those after you get the basics down.
i found the following examples to be useful in improving realistic generations, but is still useful for artistic generations too (i'll highlight them out in the example)

---

+​

"(((NSFW))), (((nudity))), (((Photo-realism:1.2))), (((Nikon Z9))), (((DSLR))), (((8k))), (((masterpiece:1.2))), (((extreme detail:1.2))), (((realistic:1.2))), (((ultra high quality))), (((absurdres))), (((HDR))), (((perfect))), ​

1girl, (((skinny))), (((wide hips))), (((smile:1.3))), (((freckles:1.3))), (((beautiful face))),(((realistic eyes))),(((perfect detailed face:1.5))),(((perfect detailed body:1.5))),(((perfect detailed eyes:1.5))),(((proper anatomy))), perfect hands, perfect fingers, perfect feet.........​

​

-​

(lowres), (worst quality), (low quality), (fuze:1.4), (bad anatomy), cartoon, anime, comic, drawing, illustration, bad hands, bad face, text, error, missing fingers, extra digit, fewer digits, extra hands, extra arms, big ears, cropped, worst quality, low quality, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name, bad skin........​

---

so the curved brackets , and the numbers are supposed to add emphasis on these specific prompts and prioritizes them over the other prompts in the field. higher numbers and more brackets adds more strength to the particular prompt.

in all honesty, i still haven't worked out the correct balance so i play around with the brackets and numbers a lot, but advise you to start low in the amount of prompts, number strength, and brackets, and work you way up with each new generation until you get what you want.

check out civitai.com - check out a few of the models, Lora's, etc. look up an image you like and check out the prompts used to make it. 4chan also had a really good guide, but be prepared to do a LOT of research and a LOT of trial and error before you get anything remotely good.

 

[randomgrizzlybear]

What's the best Ai tool for free use, I've tried looking them up but most of the good features are locked away with premium access. Also I'm a bit wary on using them, can you use loli tags without getting flagged?

 

[LostMarble]

---

+​

"(((NSFW))), (((nudity))), (((Photo-realism:1.2))), (((Nikon Z9))), (((DSLR))), (((8k))), (((masterpiece:1.2))), (((extreme detail:1.2))), (((realistic:1.2))), (((ultra high quality))), (((absurdres))), (((HDR))), (((perfect))), ​

1girl, (((skinny))), (((wide hips))), (((smile:1.3))), (((freckles:1.3))), (((beautiful face))),(((realistic eyes))),(((perfect detailed face:1.5))),(((perfect detailed body:1.5))),(((perfect detailed eyes:1.5))),(((proper anatomy))), perfect hands, perfect fingers, perfect feet.........​

​

-​

(lowres), (worst quality), (low quality), (fuze:1.4), (bad anatomy), cartoon, anime, comic, drawing, illustration, bad hands, bad face, text, error, missing fingers, extra digit, fewer digits, extra hands, extra arms, big ears, cropped, worst quality, low quality, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name, bad skin........​

---

Careful there's only so much emphasis to go around until it has a nullifing effect, just like if you use lots of tokens, each one has diminishing returns. In most software the parenthesies count as .1 weight each so (token:1.3) is the same as (((token))), and can also group tokens (nsfw nudity nikon etc:1.247).

Do you get good results with prompting like that? Ultimately that was going to be my answer to OP -- there's almost no "wrong way" to prompt since the ai has been trained on so much human language, there's just little things you can do to help the ai make the best image/text/music/whatever it can.

 

[LostMarble]

What's the best Ai tool for free use, I've looking them up but most of the good features are locked away with premium access. Also I'm a bit wary on using them, can you use loli tags without getting flagged?

Online? You're going to be paying like you said, generating images is quite a bit of compute that no one will just be giving away. Though you can get pretty far by using civitai's version of lc coins to generate stuff. Also huggingface still has free cpu inference i think, if you don't mind waiting forever to get gens. As far as loli goes, well... are you really going to have any peace of mind unless you generate that locally?

 

[randomgrizzlybear]

Online? You're going to be paying like you said, generating images is quite a bit of compute that no one will just be giving away. Though you can get pretty far by using civitai's version of lc coins to generate stuff. Also huggingface still has free cpu inference i think, if you don't mind waiting forever to get gens. As far as loli goes, well... are you really going to have any peace of mind unless you generate that locally?

I don't mind downloading software's for it, don't why I just assumed using them online was the way to go. What do you suggest for local use?

 

[Halfbreed]

I'm not very good at these, I wish there were easy ones to use online. I'm not sure what's a safe one though.

 

[LostMarble]

I don't mind downloading software's for it, don't why I just assumed using them online was the way to go. What do you suggest for local use?

Oh sorry, I just assume people looking to generate online are on mobile and don't have computers and such. Your computer is going to need some sort of nvidia gpu or you're in for a slow time. So if you got that, then I typically recommend Fooocus: for people who just want to press a button to generate images, Automatic1111 or Forge: are good all rounders and what i use most of the time, and lastly ComfyUI: for advanced workflows.

 

[randomgrizzlybear]

Oh sorry, I just assume people looking to generate online are on mobile and don't have computers and such. Your computer is going to need some sort of nvidia gpu or you're in for a slow time. So if you got that, then I typically recommend Fooocus: for people who just want to press a button to generate images, Automatic1111 or Forge: are good all rounders and what i use most of the time, and lastly ComfyUI: for advanced workflows.

Thanks for the recommends

 

[TemperTempah34260923]

Careful there's only so much emphasis to go around until it has a nullifing effect, just like if you use lots of tokens, each one has diminishing returns. In most software the parenthesies count as .1 weight each so (token:1.3) is the same as (((token))), and can also group tokens (nsfw nudity nikon etc:1.247).

very well said - take note of these points newbies

Do you get good results with prompting like that? Ultimately that was going to be my answer to OP -- there's almost no "wrong way" to prompt since the ai has been trained on so much human language, there's just little things you can do to help the ai make the best image/text/music/whatever it can.

admittedly i did not provide the best example, it was something i copied and pasted from a recent generation - i use a lot of LORA's and other embeddings and extensionions so i had to fine tune my prompts and put extra emphasis on certain tokens before i got anything half decent.

if you ask me its a steep learning curve cos once you figure out the basics, you then discover new features and settings to mess around with. i followed the guides of 4chan and started with a local install of stable diffusion 1.5 with comfyUI, and slowly learnt more as i went along. i still haven't figured out how to properly inpaint, but at least i still have photoshop.

i have no programming or artistic experience, but i like learning new things so its been an interesting journey.

 

[User_57059]

in my experience, the negative prompt field is just as important as the positive one.
you may have noticed there's also important modifiers , specifically the use of brackets and numbers next to a prompt.
LORA's and other add ons are extreamly helpful so i advise you to learn about those after you get the basics down.
i found the following examples to be useful in improving realistic generations, but is still useful for artistic generations too (i'll highlight them out in the example)

---

+​

"(((NSFW))), (((nudity))), (((Photo-realism:1.2))), (((Nikon Z9))), (((DSLR))), (((8k))), (((masterpiece:1.2))), (((extreme detail:1.2))), (((realistic:1.2))), (((ultra high quality))), (((absurdres))), (((HDR))), (((perfect))), ​

1girl, (((skinny))), (((wide hips))), (((smile:1.3))), (((freckles:1.3))), (((beautiful face))),(((realistic eyes))),(((perfect detailed face:1.5))),(((perfect detailed body:1.5))),(((perfect detailed eyes:1.5))),(((proper anatomy))), perfect hands, perfect fingers, perfect feet.........​

​

-​

(lowres), (worst quality), (low quality), (fuze:1.4), (bad anatomy), cartoon, anime, comic, drawing, illustration, bad hands, bad face, text, error, missing fingers, extra digit, fewer digits, extra hands, extra arms, big ears, cropped, worst quality, low quality, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name, bad skin........​

---

so the curved brackets , and the numbers are supposed to add emphasis on these specific prompts and prioritizes them over the other prompts in the field. higher numbers and more brackets adds more strength to the particular prompt.

in all honesty, i still haven't worked out the correct balance so i play around with the brackets and numbers a lot, but advise you to start low in the amount of prompts, number strength, and brackets, and work you way up with each new generation until you get what you want.

check out civitai.com - check out a few of the models, Lora's, etc. look up an image you like and check out the prompts used to make it. 4chan also had a really good guide, but be prepared to do a LOT of research and a LOT of trial and error before you get anything remotely good.

Thank you very much tempertempah for your helpful comments , really interesting I will try your prompts. You are rigth, it will be a lot research and trial and error. Would be really good to have a guideline for prompts and combinatinons and results, like a wiki only for prompts.
Thanks again also for the link I will check

 

[User_57059]

What's the best Ai tool for free use, I've tried looking them up but most of the good features are locked away with premium access. Also I'm a bit wary on using them, can you use loli tags without getting flagged?

Hello random, that is a good qeustion, I am looking also for a good tool which is safe to use. I will try the link what was already recommended from tempertempah.

 

[User_57059]

Careful there's only so much emphasis to go around until it has a nullifing effect, just like if you use lots of tokens, each one has diminishing returns. In most software the parenthesies count as .1 weight each so (token:1.3) is the same as (((token))), and can also group tokens (nsfw nudity nikon etc:1.247).

Do you get good results with prompting like that? Ultimately that was going to be my answer to OP -- there's almost no "wrong way" to prompt since the ai has been trained on so much human language, there's just little things you can do to help the ai make the best image/text/music/whatever it can.

Thank you very much for the additional hint, I have to learn a lot to undestand the logic behind

 

[SlickBrick]

Oh sorry, I just assume people looking to generate online are on mobile and don't have computers and such. Your computer is going to need some sort of nvidia gpu or you're in for a slow time. So if you got that, then I typically recommend Fooocus: for people who just want to press a button to generate images, Automatic1111 or Forge: are good all rounders and what i use most of the time, and lastly ComfyUI: for advanced workflows.

ComfyUI all the way