Fan Art AI Netorase Phone (Le Stag) - Fan Art Thread

[vampz]

Yep, that's pretty much it. It just takes time and practice to see what prompts work best, what LoRAs work for specific results, what art styles mix best for a specific look, what aspect works for a specific scene (and how it changes the scene with the exact same prompt) and what is simply unachievable in Civitai.

That seems to be true
Now I'm starting to wander how to create a new character from scratch (as in, create the look without remixing existing images)
Or at least, how to take an existing image and modify how the character looks, so it's not just copying someone else's work.......

 

[HoM11]

That seems to be true
Now I'm starting to wander how to create a new character from scratch (as in, create the look without remixing existing images)
Or at least, how to take an existing image and modify how the character looks, so it's not just copying someone else's work.......

The best way is to still remix someone's image with an art style that caught your fancy. And then just swap all the prompts that describe their character with the ones that describe the one you envision. They have a blonde bombshell with big breasts? Make it a goth girl with a hime cut, black lipstick, green eyes, choker, brow piercing, small breasts, etc. Or an asian redhead with sideshave. Whatever. Start with a simple pose: maybe standing, cowboy shot, simple background. Maybe sitting on a bed in a bedroom. See how you like the result. Checkpoints/LoRAs usually significantly influence the result. For instance, Laura's appearance is heavily influenced by the "All Disney" LoRA whice makes her look kinda like Vanellope from Wreck-it-Ralph. If you remove it, but leave all the prompts the same, the appearance will change.

Practical advice: don't commit things to memory. If you see an image you like, be it because of artstyle, character design, pose, specific prompt used to achieve an interesting result — save it to a collection, so that you can always return to it and use it later. See a useful LoRA? Save it to a collection. I currently have 3 collections for images: "character", "artstyle" and "pose". And a fourth one for models I liked (and I should split it).

 

[Flippie_86]

 

[Flippie_86]

That seems to be true
Now I'm starting to wander how to create a new character from scratch (as in, create the look without remixing existing images)
Or at least, how to take an existing image and modify how the character looks, so it's not just copying someone else's work.......

Yes what HoM11 said.
I almost always start by remixing picture I already like. Then change the details of the prompt as you like.
As for creating characters, its pretty much the same. Me myself have saves the prompts for any characters I like to use in a txt file, so its copy-paste instead of starting over every time.

 

[Axel34]

View attachment 4517107

View attachment 4517117

So nice

 

[vampz]

Here's how you could approach fixing the pose. Use the specific prompt "spitroast" since it seems to be what you're looking for. You can give it more weight, higher priority by writing is like this — (spitroast:1.0). The higher the number, the higher the priority. 1.0 is usually enough in my experience. "Sex from behind" is a good prompt for, well, sex from behind in general. Simpler, IMO, is often better.
Here's how I tried it.

Spoiler: Prompt

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

Let's see what this gives us.
View attachment 4516733

That's already good. Let's use "remix with seed" to use this result, but change the prompt a bit so that instead of a guy on the right there's another dick from a hole.

Spoiler: Second attempt prompt

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

And voilà:
View attachment 4516748

EDIT: yes, forgot to mention the change to landscape aspect, just as nickohoma pointed out.

Well, it took a few attempts, but I finally got what I was looking for

Spoiler

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

Thank you very much for the help
And here is a little extra thank you from the heroine of our adventures

Spoiler

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

 

[vampz]

BTW guys, I realized I had another question, somewhat more advanced
Say I'm generating a specific character A, and I like the overall look for them
Now I also generate a different character B separately
How would you approach trying to "import" both characters and generate images where both appear?
For this, assume both characters use the same model and configuration, as otherwise it would change the look of the character generated with a different config

 

[HoM11]

BTW guys, I realized I had another question, somewhat more advanced
Say I'm generating a specific character A, and I like the overall look for them
Now I also generate a different character B separately
How would you approach trying to "import" both characters and generate images where both appear?
For this, assume both characters use the same model and configuration, as otherwise it would change the look of the character generated with a different config

I asked a similar question when I started with Civitai and this is pretty much the answer. Unfortunately, Civitai has a tendency to mix character traits. Say, you wrote that one girl should have short blonde hair and the other long black hair? It can (and most likely will) make both of them have long hair, or both have blond hair, or swap their hairstyles, or something. There is no reliable way around it without using local clients. That's why I quickly gave up on trying to have Laura and Tara in the same picture (and they have pretty distinct features). It does it even with male/female pairs albeit to a slightly lesser extent (most of my pictures have "face out of frame" for men to avoid mixing features with girls).

The only exception to this seems to be using some well recognized characters that the AI (I think, the checkpoint specifically) knows. I did that recently with Lusamine and Lillie from pokemon. And even though the results overall were pretty consistent, AI still sometimes mixed the hairstyles together.

 

[Flippie_86]

Yeah, unfortunately it's difficult with AI generation in general to have multiple characters exactly as you prompted them. It takes the features described and applies them most logically according to the reference data.

This is why having Laura and Tara together, for example, is difficult. There are a lot more darker skinned girls with black hair than there are with pink hair. So when prompting for a girl with a tan, it gives the girl with the black hair the tan.



Only way to really solve this is to run AI generation locally and use tools like inpainting.

 

[vampz]

Yeah, unfortunately it's difficult with AI generation in general to have multiple characters exactly as you prompted them. It takes the features described and applies them most logically according to the reference data.

This is why having Laura and Tara together, for example, is difficult. There are a lot more darker skinned girls with black hair than there are with pink hair. So when prompting for a girl with a tan, it gives the girl with the black hair the tan.
View attachment 4518571


Only way to really solve this is to run AI generation locally and use tools like inpainting.

I guess it means it's going to be time for me to learn how to do it all locally
Any help on a good starting point would be seriously appreciated

 

[Flippie_86]

I guess it means it's going to be time for me to learn how to do it all locally
Any help on a good starting point would be seriously appreciated

I started locally with Automatic1111
This playlist helped me out a lot to start:

You must be registered to see the links

Important note, you do need a fairly decent graphics card to be able to generate AI locally.

 

[vampz]

I started locally with Automatic1111
This playlist helped me out a lot to start:

You must be registered to see the links

Important note, you do need a fairly decent graphics card to be able to generate AI locally.

Thanks
I actually found their channel by searching on google, and installed ForgeUI with the models I've been using from CivitAI

I have a 3080, so it's able to generate images rather quickly
Now I just have to figure out the prompts to get the results I want......

On a side note, do you know if it's possible to get the same result on local generation, as in Civit?
I took an image I made, used the exact same models, LoRas etc, same seed, same prompts, every setting is identical, but the image isn't quite the same, and I'm not sure how to nudge the AI into giving a closer result.......

 

[nickohoma]

Thanks
I actually found their channel by searching on google, and installed ForgeUI with the models I've been using from CivitAI

I have a 3080, so it's able to generate images rather quickly
Now I just have to figure out the prompts to get the results I want......

On a side note, do you know if it's possible to get the same result on local generation, as in Civit?
I took an image I made, used the exact same models, LoRas etc, same seed, same prompts, every setting is identical, but the image isn't quite the same, and I'm not sure how to nudge the AI into giving a closer result.......

AFAIK, you won't be able to get local gens exactly the same as CivitAI. CivitAI automatically applies an embedding to every image it creates to try to prevent underage stuff. Last I checked, that embedding wasn't available for download, so you won't be able to run it locally.

I always add "child" to my negative prompt to accomplish the same thing, but it's not identical to the CivitAI automatic embedding. If you don't add "child" to your negative though, sometimes characters are kind of puffy-faced chibi-ish.

Also if you're running Forge, make sure you're calling your LoRA's out in your prompt. They'll look like "<lora:Smooth Anime Style LoRA XL:1.0>, <lora:Expressive_H-000001:0.5>", where the numbers after the colons are the LoRA strength.

Also make sure you're using a legit height/width for your images. 832x1216 or 1024x1024 or 1216x832 are most common. Forge might default to 512x512 which will suck.

 

[Flippie_86]

On a side note, do you know if it's possible to get the same result on local generation, as in Civit?

I also get different results on Civitai then I do locally.
think there must be some hidden negative prompts in Civitai or something.

What helped me get closer to the civitai results was adding things like, kids, children, shota, loli, in the negative prompt

 

[vampz]

Hi guys
I didn't want to derail this thread, since it's focused on this game and it's art style
I realized I couldn't really replicate it, nor did I want to create a simple clone of this game
However, I have created a thread here with some artwork of the two characters I have for now (my versions of Laura and Ashley), and I would love it if you could stop by and give your feedback, and if you want I could share more of what I have prepared thus far

Special thanks to Le Stag for the inspiration, I wouldn't have done any of it without you and your wonderful game

 

[Gloub]

Sooooo ...
a little birdy told that 'till the next update, Laura and her boyfriend might have been travelling trought Europe

Spoiler

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

Spoiler

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

 

[player two]

nickohoma - I've been catching up on the all advice in this thread regarding Automatic1111 vs Comfy, which checkpoints & lora to use, etc. I've been making copious notes so after I upgrade to the latest Fedora and Mint Linux, I can start trying out making some image content.

IIRC reading earlier you mentioned:

• you saved your prompts within your seed PNG so when you load the PNGs, the prompts are pre-filled
• and for noobs like me, to download your posted/uploaded "seed PNGs" here in the forum as a starting point.

Would it be possible for you to upload some fresh seed PNGs in an archive file (ZIP, RAR, 7z,etc)?
(It would be nice to have a baseline PNG w/ metadata prompt of the NP's core characters.)

Why? Well, over in the Waifu Dream City sharing forum, I've identified that F95 is converting genuine PNGs into AVIF files with PNG extensions. That's resulting in lost embedded data that stored as "Stego", standard EXIF, and PNG Text chunks.

I would really appreciate it to know I'm starting with a "genuine" seed PNG image having some known working example prompts.

Thanks in advance. Cheers.

 

[nickohoma]

nickohoma - I've been catching up on the all advice in this thread regarding Automatic1111 vs Comfy, which checkpoints & lora to use, etc. I've been making copious notes so after I upgrade to the latest Fedora and Mint Linux, I can start trying out making some image content.

IIRC reading earlier you mentioned:

• you saved your prompts within your seed PNG so when you load the PNGs, the prompts are pre-filled
• and for noobs like me, to download your posted/uploaded "seed PNGs" here in the forum as a starting point.

Would it be possible for you to upload some fresh seed PNGs in an archive file (ZIP, RAR, 7z,etc)?
(It would be nice to have a baseline PNG w/ metadata prompt of the NP's core characters.)

Why? Well, over in the Waifu Dream City sharing forum, I've identified that F95 is converting genuine PNGs into AVIF files with PNG extensions. That's resulting in lost embedded data that stored as "Stego", standard EXIF, and PNG Text chunks.

I would really appreciate it to know I'm starting with a "genuine" seed PNG image having some known working example prompts.

Thanks in advance. Cheers.

Yeah it's a shame about F95 auto converting everything to AVIF. Makes it harder to share image data.

You must be registered to see the links

That is a sampling of some of the stuff I did. I separated out the stuff that uses regional sampling since those take some extra steps. And not everything there is necessarily the final versions of images I shared. Some would have gotten some inpainting/img2img work before I posted them. But it should give an idea what the prompts and settings were.

If you're looking for a particular image, let me know and I'll try to dig up the original.

 

[psypop404]

Can someone do more of tara

 

[HoM11]

Can someone do more of tara

Maaaaybe...