Fan Art Our Red String: Fan Art

[TIGER ACE]

I don't think you should take SerHawke's comment as a personal jab at you, he was merely continuing the topic discussion from the main thread. For that matter I hope you didn't delete your other post because you felt discouraged by my comment either.

No, nothing like that. I couldn't see the pic and wasnt sure if the upload worked lol. None of the images in the thread was loading for me.

Now to Version 2:
Prompt score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Lena, 1girl, 1boy, penis, fellatio, long hair, nude, oral, breasts, ass, blue eyes

I tried different training parameters, like I mentioned before, so this version is a little different. but character tags are in.


The identity is not strong, due to lack of High quality face shots. It is what it is, but it can't be helped.

Another one just with character tag:

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Lena,

For the most part other character tags ARE working by themselves, if they had a lot of images in the game lol, if it doesnt work you need to add the hair color tags and usually its enough. mystery of the neural net when it does or not.

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Holly,


score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Emma




All the scores arent necessary btw, its pony stuff but tons of people just use up to 7, you might have to check the difference yourself.
Also you dont have do rating_explicit, which asks for nsfw generation, you can use rating_safe for sfw images. rating_questionable puts it somewhere in the middle.

Also I recommend putting 3d in the negative prompt, maybe cartoon too.

Examples of Less score. and safe rating.

score_9, score_8_up, score_7_up, rating_safe, evakiss, Cindy :


score_9, score_8_up, score_7_up, rating_safe, evakiss, Allison:


people with less images dont have strong identity and needs atleast hair color to get close like Jess

score_9, score_8_up, score_7_up, rating_questionable, evakiss, evakiss, Jess, 1girl, blonde hair, tattoo :


Cherry was turned into Cherryl, because the model was generating cherries no matter what in the image, even when putting the girls in there, there would be random cherries in the pic.

score_9, score_8_up, score_7_up, rating_questionable, evakiss, Cherryl, dark-skinned female :

in summary, image was processed differently, used a different learning algorithm, also different amount of steps, and actually less training images. therefore its different.

 

[BlandChili]

You guys have to say if the artstyle got deviated or not. Also I wanted to say rating_explicit and the scores are pony tags , you dont have to do all of them some people onyl do upto 7, and you can use rating_safe to get sfw pics, rating_explicit forces nsfw, rating_questionable puts it somewhere in the middle.

These are better than the unrelated one from earlier, some of the details are perhaps a bit overblown compared to Eva's style. The eyes should perhaps be a bit smaller, the lashes are far less pronounced in Eva art.
For some reason some of the skin colour gradients seem odd now and then as well, for example your Emma there looks like she went tanning with a teacup covering half of each tit.
It also seems to prefer teeth in the overbite, not at all uncommon with Eva but she often shows to full set too and the AI seems to dislike making a partition in images that seem like they're showing both sets of teeth
It also seems to apply a blush to every single scene and sometimes it's less subtle than what Eva would use. In the same way the AI also omits highlighting the philtrum most times, except for when the lighting comes from the right angle.
I made a small collab for reference here:

Pay attention to the lashes, the teeth, the philtrum and the proportion of the eyes. Compare for example the example Holly with your BJ one, her nose is a bit smaller and her eyes are a bit higher on her face compared to her mouth-nose ratio (though some of it could be the angle).
The AI also sometimes makes certain faces rounder or slimmer. Emma has a more round face and Lena most often has a more slim face, though it is a bit inconsistent since Eva traces and the same character does look a bit different from image to image here and there.

 

[AtipTheFirst]

No, nothing like that. I couldn't see the pic and wasnt sure if the upload worked lol. None of the images in the thread was loading for me.

Now to Version 2: View attachment 3526895
Prompt score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Lena, 1girl, 1boy, penis, fellatio, long hair, nude, oral, breasts, ass, blue eyes

I tried different training parameters, like I mentioned before, so this version is a little different. but character tags are in.


The identity is not strong, due to lack of High quality face shots. It is what it is, but it can't be helped.

Another one just with character tag: View attachment 3526915

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Lena,

For the most part other character tags ARE working by themselves, if they had a lot of images in the game lol, if it doesnt work you need to add the hair color tags and usually its enough. mystery of the neural net when it does or not.

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Holly,
View attachment 3526925

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, evakiss, Emma

View attachment 3526929


All the scores arent necessary btw, its pony stuff but tons of people just use up to 7, you might have to check the difference yourself.
Also you dont have do rating_explicit, which asks for nsfw generation, you can use rating_safe for sfw images. rating_questionable puts it somewhere in the middle.

Also I recommend putting 3d in the negative prompt, maybe cartoon too.

Examples of Less score. and safe rating.

score_9, score_8_up, score_7_up, rating_safe, evakiss, Cindy : View attachment 3526934


score_9, score_8_up, score_7_up, rating_safe, evakiss, Allison: View attachment 3526936


people with less images dont have strong identity and needs atleast hair color to get close like Jess

score_9, score_8_up, score_7_up, rating_questionable, evakiss, evakiss, Jess, 1girl, blonde hair, tattoo : View attachment 3526970


Cherry was turned into Cherryl, because the model was generating cherries no matter what in the image, even when putting the girls in there, there would be random cherries in the pic.

score_9, score_8_up, score_7_up, rating_questionable, evakiss, Cherryl, dark-skinned female : View attachment 3526984

in summary, image was processed differently, used a different learning algorithm, also different amount of steps, and actually less training images. therefore its different. View attachment 3527006

Has to be said, really love the work you've put in to this. Never really played with Stable diffusion and this made me have a go and the model is great to use (and look forward to this updated version). Can I ask though, you said you used AI to generate the tags - how is that done, another tool (I'm using forge to generate)?

 

[TIGER ACE]

Spoiler

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

Basically, I havent tested it a lot, I think the colors are stronger than Original, and the prompts were basic, just 2 tags, you have try complex combination, the more details you give the less it makes it up, also the images were generated low strengh, not that I am saying its good, part of making loras is learning stable diffusion it self, I think this training gave model a lot more freedom, I basically tried some things to get better result but it might have made things more fluid during training. Nobody really know the exact parameters, even if you follow guide, you never get the same result.

The lines are too sharp I think because of the image processing I did, in a hope of getting better quality.

 

[BlandChili]

Basically, I havent tested it a lot, I think the colors are stronger than Original, and the prompts were basic, just 2 tags, you have try complex combination, the more details you give the less it makes it up, also the images were generated low strengh, not that I am saying its good, part of making loras is learning stable diffusion it self, I think this training gave model a lot more freedom, I basically tried some things to get better result but it might have made things more fluid during training. Nobody really know the exact parameters, even if you follow guide, you never get the same result.

And it has come impressively far. What it does do well is replicate the colour, background effects and lighting. Getting the characters right has always been what AI needed training the most so it makes sense that's where remaining hiccups stick. Keep at it.

 

[TIGER ACE]

Something with base SDXL, it fucking loves parted lips and teeth, it just refuses not to do it lol, well remember to put it in negative prompt: parted lips, teeth

 

[raining]

Great job! I just tried some prompts got from here. I am very satisfied with some results.

 

[jimul18]

View attachment 3527308

Something with base SDXL, it fucking loves parted lips and teeth, it just refuses not to do it lol, well remember to put it in negative prompt: parted lips, teeth

View attachment 3527331 View attachment 3527347 View attachment 3527373 View attachment 3527380 View attachment 3527383 View attachment 3527308 View attachment 3527329 View attachment 3527330 View attachment 3527392 View attachment 3527408 View attachment 3527412 View attachment 3527417


View attachment 3527427

DAMN! Have you updated the version on Civit as well?

 

[TIGER ACE]

DAMN! Have you updated the version on Civit as well?

I just did

You must be registered to see the links

BUT not sure how usable it will be for others, as I said, there was a issue with the training, basically, I didnt resize the images myself. I will just explain..
I dont think the lora itself is too bad, but there is issues, it needs 2-3 passes to make the example images,
I upscaled all the images to 4k and put in for training, during the training, the images are resized automatically, usually you dont need to resize at all. its fine. But idk what happened.
so while the original is this.


After auto resize, what the model trained with is this instead:



So the problem, too many dotted lines, lines that were clear becamee dotted and generated image feels too sharp if you dont retouch it. the smoothness is way less. Its fine if you inpaint it technically it is still better trained, more steps, more computation power during training, bigger dim, but quality without inpaining or upscaling is inferior. I dont use forge anymore, so maybe I will upload it and you can see if hires fix, fixes it auto. you can see it in my example images many dotted lines instead of clear definitions

 

[pokiehl]

I am using AutismPony and added the Lora to the correct folder, but it still gives me very different outputs compared to Evas, any tips?

 

[Chemish]

Im not taking the blame for what happened in the other thread,people started to post AI way before i did it,TIGER ACE first post was like one week ago.


This is what i was saying,EK is not going to use AI so why be annyoing about it in the main thread?

I don’t think anyone reasonable would blame you. I made one comment about AI in the main thread, then I learned Eva’s not in favor of it, so that settles it for me.

 

[Chemish]

I am using AutismPony and added the Lora to the correct folder, but it still gives me very different outputs compared to Evas, any tips?

Here I am wondering who names these programs.

 

[TIGER ACE]

I am using AutismPony and added the Lora to the correct folder, but it still gives me very different outputs compared to Evas, any tips?

you have to activate it, depending on the ui you are using its different, i think both are drag down from the loras, and than use the prompt, template from score 9 to evakiss. you have to tell us what you are using.

 

[pokiehl]

you have to activate it, depending on the ui you are using its different, i think both are drag down from the loras, and than use the prompt, template from score 9 to evakiss. you have to tell us what you are using.

Could you point me where do I activate it, please?

 

[TIGER ACE]

Could you point me where do I activate it, please?

I dont use that, once selected it should work with triggerword, EvaKiss, you didnt say what it did generate so I dont know if its a clip issue but clip skip should be set to 2 for pony.Also Pony XL has their own vae, you can download it from Pony XL 6 page on civitai,put that in the vae folder and select that vae instead, it could be that.

and 1 more thing is you can put this in the prompt to activate the lora : "<lora:EvaKiss:0.8>" inside quote part

 

[raining]

Do not click, if you don't like futa stuff.

Spoiler: Warning about content

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

 

[seaoflove]

Do not click, if you don't like futa stuff.

Spoiler: Warning about content

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

Back before they met in person Lena wrote a fan letter to Holly. After a week Lena received a letter back from Holly which she still has to this day. Lena requested for Holly to write her in as a character in one of her books. Holly responded back to Lena that she has always been a character in her stories. It took Lena a little while to figure out what Holly meant. So Lena grew a big fat cock and fucked Holly hard and fast like a jackhammer!

 

[Arata20]

Do not click, if you don't like futa stuff.

Spoiler: Warning about content

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

Gorgeous art. Indistinguishable from EvaKiss's drawings, except for futa Lena.

 

[DeemLeem12]

What in tarnation? The original thread got blocked because people are posting art there and because that thread is for "discussing Our Red String"? Is that mod retarded? The art postings are pertinent to the discussion, and adds to the discourse given that the game is a VISUAL NOVEL. This is nonsensical policing, how the fuck does this make that thread better? Telling people what to do despite not being harmful. Those posts actually made discussions there more vibrant given the drought of content and the otherwise gloomy mood of people around such lack of update. Did that mod not read the discussions around the art? Holy shit they need to get a hobby instead of doing unnecessary things like that.

And this is coming from someone who never posted any of those art, and just enjoyed reading and discussing the characters or plot of the game pertinent to the art.

 

[BlandChili]

I disagree. I found the art postings in the main thread very annoying exactly because we have this space for it. It wasn't so much a problem when people were sharing a fan-work or two once in a while, but it stopped really being about Our Red String once several pages in a row became only discussions about fan-art, that's off topic enough that I think it warrants policing so the rest of the thread discussions wasn't (isn't hopefully soon) drowned out by it.