- Feb 7, 2020
- 82
- 645
This is very good. You might want to try getting in touch with Eva Kiss on Patreon or Discord. Maybe nothing comes of it, but it might help speed up the process.
- Mar 10, 2019
- 230
- 1,236
- Feb 7, 2020
- 82
- 645
I used the autismmixSDXL_autismmixPony
and my promts was score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, EvaKiss, 1girl, long hair, open mouth, black hair, 1boy, ass, hetero, nude, girl on top, pussy, sex from behind, nude , spread legs, bangs, teeth, cleavage, closed eyes, reverse cowgirl position, sex, smile, big black penis, huge penis, solo focus, sex, lips, uncensored, makeup, bed sheet, on stomach, lipstick, vaginal, full attractive, beautiful, very detailed, cinematic, stunning, intricate, elegant,no bra, no panties, tongue out, dark-skinned male, interracial, veiny penis
- Mar 10, 2019
- 230
- 1,236
- Mar 10, 2019
- 230
- 1,236
I also tried "img2img" but the problem is that it doesn't keep facial proportions, so when trying to recreate the image as Emma it always creates a new face, basics are there though.
- Oct 1, 2017
- 825
- 2,619
It's amazing what you guys managed to make so far and I envy your abilities to use the ai tools. Those looks straight out of the game.
- Mar 10, 2019
- 230
- 1,236
Honestly as I suggested on the main post already this is the future of this kind of game in terms of visuals, add to that stuff that is possible through NVIDIA Omni/Studio where you can smooth out animation from a few stop motions animation images and you get pretty fluid animation.
The questions honestly is where does EK want to take this game/her work. Not to start another milking debate, but currently I don't have much hope. Though some of this has the potential of patronage close to "Wild Life" i.e. almost six figures a month. You just need to adjust for some fetishes and let the money roll in.
Some questions for you good sir. It seems that faces and characters are pretty "generic/non-specific" compared to the game, and you can't for example specify "Create Emma or Holly or Ivy" right? Or would you have to train individual models/loras for individual characters?
I have spend just about one hour trying stuff and I have to say...amazing work dude!
- Jan 9, 2019
- 119
- 344
Yes, its not trained on the characters, I could have added the tags and tried but it would not do well without some really high quality face shots only, I was mostly trying to see if Pony would nail the artstyle, and I think its better than I expected. Now I think I will make another lora for the characters but doing it really well and not get generic face that just look alike is difficult. I am not an expert or anything lol but I am going to give it a try.
I will give you guys the training tags, it has the name of the chars on it, Remember to do the score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit before you copy paste the tags, the tags were ai generated too so, just edit them based on what you want. so you use it as example / template .
Last edited:
- Mar 10, 2019
- 230
- 1,236
Nope, most likely a mashup of several likely the same as Iris in GGGB based on bodytype.
- Mar 10, 2019
- 230
- 1,236
What sampling method, SD VAE and what hires upscaler has the best results? As for sampling I tried DPM++ 3M SDE and Euler A as sampling method, speed and results are fine
What are other settings you guys use? I have sampling on 150, usually batch count and size on 2 each, CFG on 6 - 8; seed -1
also TIGER ACE on what resolution was the model/lora trained; I have read that this sometimes can give better results especially in img2img if you keep it exactly the same and later upres.
And one more thing sorry to be a bother but can I use the names of the chars? Does Webforge understand this somehow if it use for example "v10_cindy1" in the prompt?
- Jan 9, 2019
- 119
- 344
The model was made for Pony XL , which is a fintune of SDXL, so you can do any of the sdxl resolution, I think the full list is somewhere, its like variation of 1024X1024. example
Since the images in the game file are 16:9 so that probably was the training resize but i have mostly been doing 1024X1024 and 832X1216
I do 60 steps usually, cfg 7, sometime increase it to 9 if I want it more artistic, Eular A people say good with Pony but I usually use DPM++ 2M SDE because thats default for Fooocus, but you can just try and see which makes the best face for you, like Eular Sgm Uniform people say is also good.
Name of the file is not taken into account only the tags are, I did create A character Lora but as I expected its incompatiable with the artstyle, Merging is problematic for me, Again I dont know how well the characters will be done if I redo the training but I do want try tomorrow maybe, I used 500 image , I plan to increase that to 1000 image, and this time tag the characters, even though I will have to manually do that. but I am not sure how to do it when there is multiple people in image but we will see.
| 1:1 | 1024x1024 |
| 4:3 | 1152x864 |
| 3:2 | 1248x832 |
| 16:9 | 1344x768 |
Since the images in the game file are 16:9 so that probably was the training resize but i have mostly been doing 1024X1024 and 832X1216
I do 60 steps usually, cfg 7, sometime increase it to 9 if I want it more artistic, Eular A people say good with Pony but I usually use DPM++ 2M SDE because thats default for Fooocus, but you can just try and see which makes the best face for you, like Eular Sgm Uniform people say is also good.
Name of the file is not taken into account only the tags are, I did create A character Lora but as I expected its incompatiable with the artstyle, Merging is problematic for me, Again I dont know how well the characters will be done if I redo the training but I do want try tomorrow maybe, I used 500 image , I plan to increase that to 1000 image, and this time tag the characters, even though I will have to manually do that. but I am not sure how to do it when there is multiple people in image but we will see.
Dudee, that’s impressive! Could you please share a step-by-step guide with us mere mortals? I’ve tried, but I honestly have no idea how to do it, even with the explanations. Haha
- Jan 9, 2019
- 119
- 344
Depends on what you are doing it on?
SO you need a ui that can run SDXL, people like Forgeui or Fooocus, chose which you like, I only use Fooocus. Its easy to install , I dont know if you got it installed.
the model is trained on Pony Xl 6, so you need to download that,
You must be registered to see the links
I am using this.Next you need my lora
You must be registered to see the links
, You just need to do 1 or 2 run with the tags that are in this thread or in the civit ai page of the lora from the example image,Once you do some you will figure it out, just keep the first part of tags always on
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_explicit, EvaKiss
Even if you dont have anything after it will generate.
Now this sounds easy or difficult based on where exactly you are at, so let me know what did you try, as you said, in the end its an artstyle lora, so you can just make your own character too, just use appropiate tags, give details about hair, hairstyle, facial feature and see what the model does.
.
Last edited:
All this generative AI is an amazing thing; however, there are two problems with it:
1) The technical problem is that you can't have multi-layered representations of images like in Photoshop. For example, when you try to animate a scene or make variations of the image, it is vital to have control over layers. You may think that these tiny changes are not a big deal, but they are. Probably, you can solve it with good prompting, but I’m not sure it's possible now.
2) The psychological problem is that as a creator, you lack the feeling that you are creating something you love. When you generate something, you know that you don’t have control over the things that are being created by AI. You may generally love the result of AI generation, but still want to slightly change a few things. Here, you’ll face issue number 1.
I personally use AI a lot in my game development. But I spent hours and days trying to create backgrounds that I wanted, re-generating them over and over again. I used ChatGPT hundreds of times, trying to get texts for the scenes based on general descriptions, and yet I had to rewrite and reread the majority of them, if not all. You slightly change your prompt and then iteratively send it to Midjourney, ChatGPT, or whatever it is, trying again and again, until you get super annoyed and settle for whatever you have.
1) The technical problem is that you can't have multi-layered representations of images like in Photoshop. For example, when you try to animate a scene or make variations of the image, it is vital to have control over layers. You may think that these tiny changes are not a big deal, but they are. Probably, you can solve it with good prompting, but I’m not sure it's possible now.
2) The psychological problem is that as a creator, you lack the feeling that you are creating something you love. When you generate something, you know that you don’t have control over the things that are being created by AI. You may generally love the result of AI generation, but still want to slightly change a few things. Here, you’ll face issue number 1.
I personally use AI a lot in my game development. But I spent hours and days trying to create backgrounds that I wanted, re-generating them over and over again. I used ChatGPT hundreds of times, trying to get texts for the scenes based on general descriptions, and yet I had to rewrite and reread the majority of them, if not all. You slightly change your prompt and then iteratively send it to Midjourney, ChatGPT, or whatever it is, trying again and again, until you get super annoyed and settle for whatever you have.
- Jan 9, 2019
- 119
- 344
You are correct ofcourse, but the models are getting better, each new ones have better quality and greater prompt adherence, even though its limited ofcourse, for advanced users there is alot of tools available, like comfy ui and other methods that gives lots of control and layers to do diffuse images together or in layers as you said.
I am lazy and also dont want so much complicated stuff but you can use inpaint / outpaint to do alot of the heavy work, you can use a complicated prompt or scenario and get a pretty bad image, as the more complex it is , usually you get less details in everything and inpaint the parts individually and you get something amazing iteratively and make it better every time, you dont have to just use the first generation, technically you can just continue to inpaint, upscale and it you can also modify and insert things with inpaining, A lot of people also uses photoshop + stable diffusion to get better result, they take the generation to photoship than img 2 img in the sdxl than go back to photoshop again and repeat until desired results,
So there is ways to get better result though ofcourse its not fully controllable by the user yet, but I do think, the difference in a year is amazing, for a person who tried Automatic1111 and uninstalled it after trying the base sd 1.5 to now actually impressed by the generations and thinking only a year and few months passes is stunning progress.
I am not one of those AI Art shills or anything, this is just a fun hobby for me, I never went to art school, I used to draw in elementary school and never really even did anything so I lost all my abilities as I grew up and never even bothered to try much, I just like using a tool to generate some stuff lol
Anyone knows why original tread "Not open for further replies"?