- Aug 1, 2022
- 2,191
- 7,256
- Aug 1, 2022
- 2,191
- 7,256
I did noticed that the ORS thread has less moderation than other threads where your posts get deleted right away the moment you are offtopic...
I'm sure it won't be long before we'll resume in the main thread our true high-minded debates on the finer points of whatever we have dreamed up for Dubai hypotheticals, Stan's Rohinhood investment returns, Ivy's 4D chess opening moves, the morality of GGGB cameos, and the plausibility of the Lola conspiracy.
- Feb 7, 2018
- 3,787
- 18,678
I have been having some success using both loras together. It did take me a while to find a balance. Not going to spam this thread with what I've generated but it is very fun to play around with combos that will likely never happen. Or we at least won't see since neither of the main characters are involved.
I personally prefer V1 but I probably just need to tinker with V2 more.
- Dec 15, 2020
- 2,188
- 4,998
I don't think any were stated directly. Since the comment mentions the fan-art spam, presumably it was locked to clean out those comments.
- Dec 15, 2020
- 2,188
- 4,998
- Jan 9, 2019
- 127
- 572
lora use v1
You must be registered to see the links
I use autismmix xl, but any ponyxl merge and base will work,
v1 is better and I would recommend it to get good result without multiple passes.
V2 imo inpaining is required to fix the dotted line and extra sharp output due what happened to the images during resize.
- Jan 9, 2019
- 127
- 572
Pony is a great model, so you should learn examples of what things are possible and use it for your generation like
before and after, bimbofication stuff like that. combine with other loras that doesnt change the artstyle,
before and after, bimbofication stuff like that. combine with other loras that doesnt change the artstyle,
Last edited:
- Dec 15, 2020
- 2,188
- 4,998
TLDR: AI can do some cool single images now, don't get me wrong. But they don't yet feel human or consistent enough to create a good scene without taking more effort than it would to just draw or trace them. There is always something off about them & generic-feeling. Really not good yet for quality storytelling.
From what I see, there are three main issues with AI art in its current capabilities for the sort of storytelling needed in a good VN:
1. Consistency: The character will always vary from image to image. Even if honed in well with good training to differentiate from other characters, the form tends to change slightly. This is jarring in short "day in a life" timespans and especially in scenes with multiple images of the same characters in different actions & angles, where the hair, clothing and physique vary slightly but noticeably from one to the next. It breaks continuity and with it immersion and suspension of disbelief needed for proper storytelling.
2. General control: A good creator/storyteller knows the proper positioning & angles they want to show for the vibe and (especially for porn) the heat of the scene, or at least can dig up reference images for such. Text to image is either too stiff or too random. This one is the easiest fix through img 2 img capabilities & some programs that allow you to use a quick rudimentary sketch for reference on positioning, though it's not perfected yet. It will probably be the soonest to be perfected by AI with new advances.
3. Facial expressions & human feel: Current facial expressions are very rudimentary. AI almost always feels very synthesized currently (of course, because it is) & characters don't feel human because of that. Improvements will be made with training for a wider range of more subtle differences. But the best short-term solution will likely come from improvements to img2img & those programs that translate rudimentary sketches to fleshed-out images, matching them with a trained character & rendering them to look both good/natural and consistent for that character. Ideally, there will be a program that detects the character's facial shape from various angles in a generated image & allows the user to manipulate the expressions shown in eyes, brow, mouth, cheeks to make subtle changes via sliders & presets, before re-rendering to match & smooth out the character in that expression. There are already a few programs that attempt this (sans the re-render), but it's very wonky/limited and only works from a few set angles when it works well at all.
For those with lower quality standards in storytelling (both from creator & consumer) current AI works great. Given the bland, blank or hackish expressions & composition in some bad Das-3d, Honey Select & Anime-drawing-style games, there is either little difference or even some improvements art-wise. For good content, it's just not there yet, though I think it may be within the next decade.
Writing-wise: It will never be there for good content. At least not within our lifetime. Far too many human intricacies. Sure, it will beat "than fux me wit yer big cox likc my bobs" quality you currently see on the average F95zone game. But the average F95zone game (those 3-4 star rated games & sometimes a bit above) feels like it was written by a sub-80-iq sheltered 12 year old with no sexual experience & little real-life human interaction.
From what I see, there are three main issues with AI art in its current capabilities for the sort of storytelling needed in a good VN:
1. Consistency: The character will always vary from image to image. Even if honed in well with good training to differentiate from other characters, the form tends to change slightly. This is jarring in short "day in a life" timespans and especially in scenes with multiple images of the same characters in different actions & angles, where the hair, clothing and physique vary slightly but noticeably from one to the next. It breaks continuity and with it immersion and suspension of disbelief needed for proper storytelling.
2. General control: A good creator/storyteller knows the proper positioning & angles they want to show for the vibe and (especially for porn) the heat of the scene, or at least can dig up reference images for such. Text to image is either too stiff or too random. This one is the easiest fix through img 2 img capabilities & some programs that allow you to use a quick rudimentary sketch for reference on positioning, though it's not perfected yet. It will probably be the soonest to be perfected by AI with new advances.
3. Facial expressions & human feel: Current facial expressions are very rudimentary. AI almost always feels very synthesized currently (of course, because it is) & characters don't feel human because of that. Improvements will be made with training for a wider range of more subtle differences. But the best short-term solution will likely come from improvements to img2img & those programs that translate rudimentary sketches to fleshed-out images, matching them with a trained character & rendering them to look both good/natural and consistent for that character. Ideally, there will be a program that detects the character's facial shape from various angles in a generated image & allows the user to manipulate the expressions shown in eyes, brow, mouth, cheeks to make subtle changes via sliders & presets, before re-rendering to match & smooth out the character in that expression. There are already a few programs that attempt this (sans the re-render), but it's very wonky/limited and only works from a few set angles when it works well at all.
For those with lower quality standards in storytelling (both from creator & consumer) current AI works great. Given the bland, blank or hackish expressions & composition in some bad Das-3d, Honey Select & Anime-drawing-style games, there is either little difference or even some improvements art-wise. For good content, it's just not there yet, though I think it may be within the next decade.
Writing-wise: It will never be there for good content. At least not within our lifetime. Far too many human intricacies. Sure, it will beat "than fux me wit yer big cox likc my bobs" quality you currently see on the average F95zone game. But the average F95zone game (those 3-4 star rated games & sometimes a bit above) feels like it was written by a sub-80-iq sheltered 12 year old with no sexual experience & little real-life human interaction.
Last edited:
- Mar 10, 2019
- 290
- 1,456
Damn... what was your prompt?Do not click, if you don't like futa stuff.
You don't have permission to view the spoiler content. Log in or register now.
- Mar 10, 2019
- 290
- 1,456
How did you achieve this? What lora did you combine yours with and what was the prompt?
- Jan 9, 2019
- 127
- 572
I dont really care for ai discussion, because people are constaly talking about whats possible today, and not the speed of improvements, SDXL is a year old tech already.
Everything I have posted was made on a 4070ti each generation taking about 30 sec and maybe another 30 sec of inpaining,
I think expecting it to beat a skilled human taking hours is silly.
And in the end you are not seeing what is possible with current tech, you are seeing consumer grade version of whats possible on a budget computer by non experts just pressing generate.
Everything I have posted was made on a 4070ti each generation taking about 30 sec and maybe another 30 sec of inpaining,
I think expecting it to beat a skilled human taking hours is silly.
And in the end you are not seeing what is possible with current tech, you are seeing consumer grade version of whats possible on a budget computer by non experts just pressing generate.
Last edited:
- Oct 1, 2017
- 825
- 2,622
Neither do the images EvaKiss shits outs. They feel lifeless as well, and there are like 5 images per sex scenes, which is abyssmal.
Last edited:
- Oct 1, 2017
- 825
- 2,622
No, it's not.
This was a great thread where people posted fanart and how to make it and then the same people who made the discussion thread get shut down came here to do the same with this one.
Nobody wants to hear you (this isn't directly aimed at you camube) bitch about NTR or bitch about people who like NTR. Grow up, get over it and get lost.