Fan Art Kingdom of Deception and Noxian Nights [fan art, ai models, waiting for updates, future plans, healthcare]

[Mamaragan]

I think depth is the best for style transfert. Canny is waayy too strong, since it's an edge processor, it translates everything, the face and hairstyle ends up being different for the generated render. Playing with the threshold of controlnet weight is key. Above 8 and it's too strong. It's also not perfect, sometimes you need to generate more pictures until you get the perfect seed.
The most annoying part of this tho, is resolution. If the resolution is not the same as the sdxl compatible resolution, you get distortions, artifacts and so on.

Should create another spoiler in firts post with links for extensions, preprocessors, diffusers and so on?

Good idea. I remember getting confused with model. SD15 and XL are very different and are not interchangeable.

Spoiler: SABIA

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

 

[TERES]

+1

 

[KGARrokh]

We can draw new images with Scribble module) Just need some experience in this)

Spoiler: WARNING! SHOCKING CONTENT! NOT FOR ANYONE! NO LEWDS! ONLY FUN! LEWDS WILL BE LATER!

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

 

[Magnelo]

Wow, controlnet reference model is amazing!

Spoiler: Results

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

since its a bit confusing for me, how did you make controlnet to work? On civit if you want to use it, you get this info and in the end you can't really use it

 

[KGARrokh]

since its a bit confusing for me, how did you make controlnet to work? On civit if you want to use it, you get this info and in the end you can't really use it

ControlNet is now available for use on CivitAI, you can only download some ControlNet models for your UI. We use ControlNet in ForgeUI. I think they add ControlNet support in future.

 

[KGARrokh]

"Depth" is indead the best module to recreate pose or scene, while reference "module" is good way to approximate scene.

Spoiler: Lewds from ControlNet tests:

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

Spoiler: Small bonus:

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

 

[big_shl0m0]

"Depth" is indead the best module to recreate pose or scene, while reference "module" is good way to approximate scene.

Spoiler: Lewds from ControlNet tests:

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

Spoiler: Small bonus:

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

Amazing results!
Like I always said, Sabia deserves double anal, and there's clearly not enough of it.

 

[Mamaragan]

The power of depth in controlnet


Control weight: 0.5 (at first I put it at 0.7, but the generated picture constantly took the haircut of the controlnet reference image.
Control mode: balanced. You'll probably want to let it on balanced, unless you have issues with the generated picture.


For this one, the weight is at 0.6

 

[Alicja Erberg]

"Depth" is indead the best module to recreate pose or scene, while reference "module" is good way to approximate scene.

Spoiler: Lewds from ControlNet tests:

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

Spoiler: Small bonus:

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

Can you attach original controlnet pictures as well?

 

[Enigmanic]

Amazing results!
Like I always said, Sabia deserves double anal, and there's clearly not enough of it.

No, absolutely not. Double vaginal and double anal are disgusting. This game is clearly aimed at heterosexual audiences. We don't need to see two guys rubbing their dicks together.

 

[CourierNPC]

The new method of generating images at CivitAI produces better results, especially with embedding. There are a few quirks especially on kissing poses, of course.

Spoiler: Some I have this week

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

 

[Enigmanic]

Spoiler: More practice with a once-arrogant elf.

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

 

[KGARrokh]

Spoiler: Some videos generated with Deforum, but nothing special:

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

Deforum more for creativity videos, not for scenes.

 

[KGARrokh]

Spoiler: More practice with a once-arrogant elf.

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

Your images looks overfitted, this can be caused due to high strength or because of using to many LoRA with again high strength.

Spoiler: First case demonstration:

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

Look like you use strength >= 1, try to change to 0.7.

Spoiler: Prompts and settings are built in this image, drop this in PNG info tab:

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

 

[Mamaragan]

Your images looks overfitted, this can be caused due to high strength or because of using to many LoRA with again high strength.

Spoiler: First case demonstration:

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

From the exif data, it might be because of the steps from the sampler he choses to use ( Steps: 50, Sampler: DPM++ 2M Karras, ) The weight of the lora is at 1. The over saturation is because of the sampler I guess.

I suggest him to use "euler a" at a step between 20 and 30.
Also, always use the tags score_9, score_8_up, score_7_up before any prompt, it's a requirement with PonyXL

 

[KGARrokh]

From the exif data, it might be because of the steps from the sampler he choses to use ( Steps: 50, Sampler: DPM++ 2M Karras, ) The weight of the lora is at 1. The over saturation is because of the sampler I guess.

I suggest him to use "euler a" at a step between 20 and 30.
Also, always use the tags score_9, score_8_up, score_7_up before any prompt, it's a requirement with PonyXL

Spoiler: Strength affected more then number of steps:

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

I think we just blind guessing right now) Enigmatic can you show us your txt2img settings or try this settings and prompt in your UI?

Spoiler: Example:

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

Also maybe you use different LoRA model?

Spoiler: I use this model:

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

 

[Enigmanic]

Enigmatic can you show us your txt2img settings or try this settings and prompt in your UI?

Spoiler: Example:

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

 

[KGARrokh]

Can you attach original controlnet pictures as well?

Sorry i misinterpret your message)

Spoiler: This one was used for first five images with Ylva (artist:Firebrand):

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

My first attempt was to recreate this image with Neve, with ears pull, not hairs, but i failed)

Spoiler: This was used for Neve (artist:Nomo):

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

Spoiler: Source for double anal (artist: Fenrir's Revenge):

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

Spoiler: Another one was posted earlier by Mamaragan, i don't know artist name:

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

Spoiler: For last two i used this screenshot from porn with Cathy Heaven (if you want to know actress name):

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

Thats all. Maply was generated without ControlNet assistance.

Depth module works a lot better with 3d renders, rather then porn screenshots.

 

[Mamaragan]

View attachment 3736370

Before any prompt, put score_9, score_8_up, score_7_up in the positive prompt and score_6, score_5, score_4 in the negative prompt, again, before anything.

This is the image I generated with your prompt and seed :
This with adding the score_up prompt from above :
And this one by changing the sampler with euler a and steps of 30

Same, but adding Adetailer (face yolo 8s.pt detection threshold 0.7 and mediapipe_face_mesh_eyes_only at 0.7 as well :

The image is now better

 

[KGARrokh]

View attachment 3736370

This is your problem: <lora:Kingdom_Of_Deception_characters_v.3ch__PDXL_6__r1:1.0>.
Just lower the strength like this: <lora:Kingdom_Of_Deception_characters_v.3ch__PDXL_6__r1:0.7>,
or try other values less then 1 (0.7 is recommended).

Mamaragan also right, this scores and dataset tags is crucial for any Pony model.

And one more thing, you use wrong description for Neve:

Spoiler: Descriptions for all characters for v.3ch:

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

Spoiler: You can find info about usefull tags here:

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