[Stable Diffusion] Prompt Sharing and Learning Thread

[Synalon]

I've not spent any time trying to fix the various faults in these, all are generated with the same simple prompt, but different models.
No idea if this is the sort of "scene" you're after, maybe i've been shooting at the wrong goal.

View attachment 3259916
View attachment 3259915 View attachment 3259914 View attachment 3259913

Spoiler: Prompt

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

Something like this yeah, without the crowd in sight and with minimal people to have to fix.
Usually cabaret is on a stage though and not a ballroom, and with some mind of stagelights, spotlights.

I had something like this video in mind.

You must be registered to see the links

 

[me3]

Something like this yeah, without the crowd in sight and with minimal people to have to fix.
Usually cabaret is on a stage though and not a ballroom, and with some mind of stagelights, spotlights.

I had something like this video in mind.

You must be registered to see the links

I suspect the AI is a bit unfamiliar with what would be considered "on stage", it's probably too a specific/technical term.
Best option might be to find relatively clean images of what you'd want, cut out things if need be, even pasting things together can work.
Then use those images to create a background.
I doubt depth mapping will work (at least in terms of flexibility), so maybe something like IPadapter masking/layering for background or you could try doing latent composition. Either by using the images as "seeds" for a generation or it might work converting the image directly if it's good/clean enough. Running the composition through sampling should let you control some of the style and let things "melt" together
At least with those methods you can force the background and direct it more like you want

 

[Mr-Fox]

I suspect the AI is a bit unfamiliar with what would be considered "on stage", it's probably too a specific/technical term.
Best option might be to find relatively clean images of what you'd want, cut out things if need be, even pasting things together can work.
Then use those images to create a background.
I doubt depth mapping will work (at least in terms of flexibility), so maybe something like IPadapter masking/layering for background or you could try doing latent composition. Either by using the images as "seeds" for a generation or it might work converting the image directly if it's good/clean enough. Running the composition through sampling should let you control some of the style and let things "melt" together
At least with those methods you can force the background and direct it more like you want

Have you tried latent composition? It sounds interesting. I have seen it here and there at a glance but not tried it.

 

[Sepheyer]

Have you tried latent composition? It sounds interesting. I have seen it here and there at a glance but not tried it.

I tried those, and eventually switched to i2i as being far superior.

The issue with latent compose is say a coconut in a background image gets developed into a random face in addition to the foreground image also being developed into the face you actually want. And then you have all these extra faces and people polluting your image. So, there is quite a high percent of refuse.

I attached a sample for latent compose. These are made with foreground latent of a woman onto latent of a beach, then developed together. The problem you get lucky very very infrequently (comparatively), and furthermore, some backgrounds result in success ratio being something close to 0.0001% success.

So, i ran with that workflow for a while, enough to embrace i2i as the superior workflow.

PS. I came to a conclusion that I am better off separately generating foreground pixels and background, composing them together as pixels, then running i2i on it to bake the look. That's how much i2i is better than latent compose. Also, in this case your pixels can be waaay under developed and stay at ~5 - 8 steps rather than 20.

 

[Mr-Fox]

I tried those, and eventually switched to i2i as being far superior.

The issue with latent compose is say a coconut of a background image gets developed into a face in addition to the foreground image also being developed into a face. And then you have all these extra faces and people polluting your image. So, there is quite a high percent of refuse.

I attached a sample for latent compose. These are made with foreground latent of a woman onto latent of a beach, then developed together. The problem you get lucky very very infrequently (comparatively), and furthermore, some backgrounds result in success ratio being something close to 0.0001% success.

So, i run with that workflow for a while, enough to embrace i2i as the superior workflow.

PS. I came to a conclusion that I am better off separately generating foreground pixels and background, composing them together as pixels, then running i2i on it to bake the look. That's how much i2i is better than latent compose. Also, in this case your pixels can be waaay under developed and stay at ~5 - 8 steps rather than 20.

View attachment 3262341

So.. The truth finally comes out. You are cheating..

 

[Mr-Fox]

was just teasing in case someone is wondering . .

Seph has his own distinct original style and always produce jaw dropping stuff.

 

[Jimwalrus]

was just teasing in case someone is wondering . .

Seph has his own distinct original style and always produce jaw dropping stuff.

Indeed. And one can spot his 'style' a mile away. Usually because her hips & tits are big enough to be visible from that distance!

 

[bigferrari2022]

Looking for help merging two faces to create one, reusable face.
I tried to train dreambooth with about 30 images, 15 of each face, but i don't have enough vram and it crashes. I have a 1660ti on my laptop.

I want to use two faces.... of women I know, and merge them together to create a totally new face.. I don't want to use either of their faces for obvious reasons...

I could maybe try and create a lora but i'm having difficulty.

Any tips on merging two faces, or training two faces.

I currently use Reactor with great results, but want a unique female face.

 

[sharlotte]

Have you tried to generate images including both in the generation (assuming they are somewhat famous and recognized by the AI): (Margot Robbie:0.6, Scarlett Johansson:0.7)
Prompts included in this quickly generated image:

 

[bigferrari2022]

Have you tried to generate images including both in the generation (assuming they are somewhat famous and recognized by the AI): (Margot Robbie:0.6, Scarlett Johansson:0.7)
Prompts included in this quickly generated image:
View attachment 3265668

Ahh thanks for the tip, but their not famous.. i'm familiar with the method you mentioned. I guess i need a better video card so I can run dreambooth...

 

[Mr-Fox]

Ahh thanks for the tip, but their not famous.. i'm familiar with the method you mentioned. I guess i need a better video card so I can run dreambooth...

An alternative is to use ip adapter in controlnet. You would need to use one "unit" for each and then set the control weight and starting control step etc to mix them. i have no idea if it works well but it's worth a try.

 

[bigferrari2022]

An alternative is to use ip adapter in controlnet. You would need to use one "unit" for each and then set the control weight and starting control step etc to mix them.

alright i'll give this a shot. thanks for the info.

 

[me3]

Have you tried latent composition? It sounds interesting. I have seen it here and there at a glance but not tried it.

Not recently, but it works fairly well.
However you need to be using prompts that "fit" together and just do a very small separate sampling before putting things together. So you won't really know how element look until you're done.
For comfyui users this can serve as

You must be registered to see the links

, not sure how you'd do this in a1111, workflow is in the images. For those that don't use comfy you can upload the image to

You must be registered to see the links

, scroll down about half the page and you should see a box mentioning "drag and drop ... png or json"

 

[Jimwalrus]

Looking for help merging two faces to create one, reusable face.
I tried to train dreambooth with about 30 images, 15 of each face, but i don't have enough vram and it crashes. I have a 1660ti on my laptop.

I want to use two faces.... of women I know, and merge them together to create a totally new face.. I don't want to use either of their faces for obvious reasons...

I could maybe try and create a lora but i'm having difficulty.

Any tips on merging two faces, or training two faces.

I currently use Reactor with great results, but want a unique female face.

If you've created TIs for each face, then you can use the Embedding Inspector to combine them at whatever relative strengths you wish (+ a dash of, for instance, Ariana Grande if you like).
If you've created LoRAs for each of them, I believe Kohya_ss has a LoRA combination tool.

 

[bigferrari2022]

If you've created TIs for each face, then you can use the Embedding Inspector to combine them at whatever relative strengths you wish (+ a dash of, for instance, Ariana Grande if you like).
If you've created LoRAs for each of them, I believe Kohya_ss has a LoRA combination tool.

Thanks. I'm not that advanced yet. I used img2img, and slowly blended them together for a decent result

 

[me3]

Just to followup my previous post
This just a VERY quick mockup test following the layout from the link in that post and sticking with the cabaret interest that started it
I should have given a bit more thought to the model i picked and the prompts were just "stage at moulin rouge" and "female cabaret dancer at moulin rouge" with the usual quality pos and negatives. As it's just simple SD15 i did run it through a face detailing when upscaling.
There's some problems but they are by far more affected by lacks of prompting and tweaking settings, than the method itself.

 

[devilkkw]

Looking for help merging two faces to create one, reusable face.
I tried to train dreambooth with about 30 images, 15 of each face, but i don't have enough vram and it crashes. I have a 1660ti on my laptop.

I want to use two faces.... of women I know, and merge them together to create a totally new face.. I don't want to use either of their faces for obvious reasons...

I could maybe try and create a lora but i'm having difficulty.

Any tips on merging two faces, or training two faces.

I currently use Reactor with great results, but want a unique female face.

Lora is better way, more controllable than embedding. what problem have you on training lora? if you expose maybe we give better help.
Also, for lora, you have 2 way to chose: train directly mixed, or train 2 model and merge them.

 

[devilkkw]

Testing all 22 sampler in CUI, same settings, changing only sampler.
As you can see, LCM sampler is really different, all other similar.

Spoiler: settings

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

Spoiler: prompt

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

Spoiler: note!

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

Spoiler: All sampler result

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

Hope this comparison is useful, sorry for not attaching workflow on image, but is all messed up with node i'm experimenting and not good for sharing at the moment.

 

[bigferrari2022]

Lora is better way, more controllable than embedding. what problem have you on training lora? if you expose maybe we give better help.
Also, for lora, you have 2 way to chose: train directly mixed, or train 2 model and merge them.

Hey thanks for the support. I don't think my video card has enough juice to train a lora.. or i'm not doing it properly..
I've watched a few walkthrough videos, but I can't seem to figure it out..

 

[Mr-Fox]

Testing all 22 sampler in CUI, same settings, changing only sampler.
As you can see, LCM sampler is really different, all other similar.

Spoiler: settings

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

Spoiler: prompt

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

Spoiler: note!

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

Spoiler: All sampler result

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

Hope this comparison is useful, sorry for not attaching workflow on image, but is all messed up with node i'm experimenting and not good for sharing at the moment.

That's some mighty fine Wendigo Erotica. A little tip or just observation, the LCM sampler will not give a good result if you are not using it with a LCM checkpoint. Also, if you use a resolution 1024 wih SD1.5 you are more likely to get conjoined twins. I would recommend to use 960x640 and then use either hiresfix or upscale in img2img with SD Upscale script. I know for a fact that you are already aware. This is only a reminder and for anyone else that might not be aware.