[Stable Diffusion] Prompt Sharing and Learning Thread

[Deleted member 5469217]

 

[Mr-Fox]

I use either Latent Couple or Regional Prompter to get multiple people in an image, works well View attachment 2556408

View attachment 2556426 View attachment 2556428

I have been trying to use regional prompter to get a full body (full character) image but the instructions on github was very lacking.

You must be registered to see the links

Could you give me some pointers?

The image(s) I want to turn into full character:

The kind of result I've gotten so far:

I'm sure I just misunderstood something and/or made a simple mistake.

 

[Sgravo05]

You need to be more specific - do you already have (or have found on Civitai) a Textual Inversion / LoRA of the celeb and want to make SD produce images that look so much like them their lawyers' palms get sweaty?

Spoiler: For example

You don't have permission to view the spoiler content. Log in or register now.

Or do you have a celeb in mind and want to create a TI/LoRA of them (in order to produce yada-yada-yada...)?

thank you, btw i dont have anything specific in mind, I’m just experimenting.

 

[Jimwalrus]

thank you, btw i dont have anything specific in mind, I’m just experimenting.

So are we all!
Assuming you've got SD fully installed locally (it's an assumption, for sure, but there are some very good tutorials and walkthroughs linked in the OP of this thread if you haven't) - get yourself to

You must be registered to see the links

and search on the tag 'Celebrity', download a TI or LoRA for the celeb you want, drop it into the relevant folder on your machine (see the Glossary) and use the LoRA or TI name in the prompt, plus any trigger words for the LoRA.

The Civitai Helper extension will be your friend once you've got loads of them downloaded!

Grab yourself several photorealistic Checkpoints. CyberRealistic, DeliberateRealistic that sort of thing. This will take a while as they are all 2GB plus, but they're worth having as the Checkpoint can be the key - an embedding may look terrible in one, perfect in another.

Recommended weights* (the ":0.8" etc bit after the name e.g. <lora:AmberHeard_v10:0.8>) are between 0.6 & 1.1 for LoRAs and 0.9 & 1.2 for TIs. This is unless you want to make very drastic changes to the person's look e.g. a cartoon version of Natalie Portman with cat ears, when you should turn the weight down to allow for more flexibility.

*Assuming no recommendation from the creator of the embedding.

Have a look on Civitai at the example images and ones that other people have created, then "be inspired by" their prompts. A good place to start if you've not done any before is to recreate an image someone else has done - try the Cameron Diaz pic I created above. Drop it into the PNG Info tab and 'Send to Txt2Img'. Provided you've downloaded and installed my Cameron Diaz TI and the PhotoMergeV1.2 checkpoint it should recreate that image.

Best of luck, you're well on your way to creating your very own imaginary sex airline of cloned celebrities...

 

[Sharinel]

I have been trying to use regional prompter to get a full body (full character) image but the instructions on github was very lacking.

You must be registered to see the links

Could you give me some pointers?

Here's how I set up Regional Prompter



And here is the prompt I used.

ADDCOMM (((daylight))), (((tropical beach))), professional majestic oil painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, photorealistic painting art by midjourney and greg Rutkowski ADDROW
Margot Robbie, (((patterned microbikini))), (defined cameltoe), (defined pokies), portrait ((Full body))) of ((stunningly attractive)) a sexy woman ((perfect feminine face)), intricate, 8k, highly detailed, (extremely detailed CG unity 8k wallpaper), ((square jaw)), (well defined jaw), (downturned lips), (thick lips:1.4), (detailed anatomy) ADDCOL
beautiful feet, bare feet

So ADDCOMM defines anything in the area 0 (the background). Why? because we have ticked 'use common prompt'
then anything after the ADDROW defines the next row, which in this case is 1+2. However because we have an ADDCOL it splits it up, anything before the ADDCOL goes into area 1, anything after goes into area 2

Couple of examples




Play about with it, obviously my feet prompt had hardly anything in it, this was just a proof of concept

 

[KingBel]

Play about with it, obviously my feet prompt had hardly anything in it, this was just a proof of concept

Been playing around with an extension called Tiled Diffusion as well.. works in pretty much the same way except that you can draw your regions instead of using the ADDCOL, ADDROW syntax

 

[Mr-Fox]

Been playing around with an extension called Tiled Diffusion as well.. works in pretty much the same way except that you can draw your regions instead of using the ADDCOL, ADDROW syntax

Yeah this sounds more like the extension I have. Thank you very much Sharinel for answering me and taking the time to explain but it seems that we are not using the same extension.

You must be registered to see the links

I think it was seph who posted the link for it a while ago. It contains Tiled Diffusion, Regional Prompt control and Tiled VAE.

 

[rayminator]

i have been test Stable Diffusion AUTOMATIC1111 and Stable Diffusion Automatic

Stable Diffusion AUTOMATIC1111 is a little slow on generating images
Stable Diffusion Automatic is a little faster on generating image

Stable Diffusion Automatic link

You must be registered to see the links

they are similar to each other

Spoiler: positive prompts

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: negative prompts

You don't have permission to view the spoiler content. Log in or register now.

 

[Mr-Fox]

I found a better tutorial for the MultiDiffusion extension I have been posting about.

You must be registered to see the links

You must be registered to see the links

 

[Mr-Fox]

Not using any extension, I'm only focusing on the prompt for now.
Apparently SD thinks that flower decorations should go with the "floralprint"..

 

[osanaiko]

i have been test Stable Diffusion AUTOMATIC1111 and Stable Diffusion Automatic

Stable Diffusion AUTOMATIC1111 is a little slow on generating images
Stable Diffusion Automatic is a little faster on generating image

Stable Diffusion Automatic link

You must be registered to see the links

they are similar to each other

Reading through the page, that looks really interesting, I'll give it a try. It's such a fast moving space right now.

Also about that image, I assume you're using the dark cookie lora. I find it intriguing how the eyes and some of the hair still has the "photo realistic" level of detail break through, while the rest is clearly DC style. nice work!

 

[rayminator]

Reading through the page, that looks really interesting, I'll give it a try. It's such a fast moving space right now.

Also about that image, I assume you're using the dark cookie lora. I find it intriguing how the eyes and some of the hair still has the "photo realistic" level of detail break through, while the rest is clearly DC style. nice work!

that's because I used 3D in the prompts

 

[sharlotte]

Apparently SD thinks that flower decorations should go with the "floralprint"..

Yeah, i've seen this before - I tried 'flowery dress' or 'dress with flower prints' and it looked like a wild garden on my result

 

[Mr-Fox]

Yeah, i've seen this before - I tried 'flowery dress' or 'dress with flower prints' and it looked like a wild garden on my result

The annoying thing is that if it weren't for that "decoration" and that random wall the image would have been perfect.
It's always something..

 

[miaouxtoo]

Using "By <artist/style>" does seem to make a difference, alongside the checkpoint/model selection ... maybe? Haha. The more stuff you have on screen, the higher the sampler might need to go as well. More testing req. I borrowed someone else's prompts to test

Details in png

Chilloutmix (20 steps as I usually do)


vs.

soapmix (30 steps I think?)

 

[sharlotte]

Trying out something different from naked chicks... feathered birds!
I used a trick I saw where you enter (from_below:1.3) in your original prompt, then use xyz, select promptS/R and list the different view you want to see in your images. Like so:


And here are some of the results:
Of course you can use the same technique with naked chicks too (and vary like (from_behind:1.3)... Useful technique.

 

[Sharinel]

So here was this mornings exercise.

The model used was Hyper V1 -

You must be registered to see the links

So I saw this picture and thought I would do something with it :-

You must be registered to see the links

It uses the prompt -
award winning photo, extremely detailed, amazing, fine detail, highly detailed woman, (extremely detailed eyes and face), full body, large breasts, (blue jeans), fitness, short blond hair, (stripes, belts), by Frank Cho and Terry Dodson, wrist watch, (dark alley, dark night), (white croptop),


Trying the exact same prompt but using the Hyper model gave me this :-


So then I decided to take out the 'by Frank and Terry' part. Prompt now was award winning full body photo, extremely detailed, amazing, fine detail, highly detailed woman, (extremely detailed eyes and face), full body, large breasts, (white croptop), (blue jeans), fitness, short blond hair, (stripes, belts), wrist watch, (dark alley, dark night),



I wasn't really liking the jeans so decided on miniskirt instead and decided to bling it up and change the dimensions to try and get full body shots. Prompt changed to :- award winning full body photo, extremely detailed, amazing, fine detail, highly detailed woman, (extremely detailed eyes and face), full body, large breasts, (intricate croptop), (intricate miniskirt), fitness, short blond hair, intricate jewelery, wrist watch, (dark alley, dark night)



Lets get rid of the alleyway and try a fashion catwalk. Prompt now is :- award winning full body photo, extremely detailed, amazing, fine detail, highly detailed woman, (catwalk), (extremely detailed eyes and face), strong jaw, high cheekbones, full body, large breasts, (intricate transparent croptop), (intricate miniskirt), fitness, short blond hair, intricate jewelery



Got a bit lucky with the background girl as A. I didn't prompt for her and B. She doesn't look like a monster, even has 5 fingers!

So that was my morning while hard at work

 

[Deleted member 5469217]

If you guys would like to participate in my poll i would really appreciate it to know what i would like to do in the future

You must be registered to see the links

 

[sharlotte]

Time for confession girls:

 

[Mr-Fox]

Time for confession girls: View attachment 2563011

Keflom!..