QSP Jack-o-Nine-Tails [v2.3] [ImperatorAugustusTertius / qwertyu12359]

[bews]

With the new model release now we can start working on the backgrounds. qwertyu12359 I think we need ur help here.

AI can produce very beautiful and detailed images now:

Spoiler

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

But it requires a detailed description. For example I can easily upscale old existing backgrounds:

Spoiler

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

But here are 2 problems:
1. Some details are lost because I didn't describe them.
2. Those old images are not perfect to begin with, so it might be better to create something completely different using some other images as a base.

So bc of this I need a detaled description of each background: what exactly should be there (all small details) and how the composition should look like (need a reference image). We can start with 2-3 backgrounds and see how it will go.

The main concern here is lore, bc I played it a long time ago and don't really remember anything. So we need someone who knows the lore to make those descriptions.

 

[qwertyu12359]

With the new model release now we can start working on the backgrounds. qwertyu12359 I think we need ur help here.

AI can produce very beautiful and detailed images now:

Spoiler

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

But it requires a detailed description. For example I can easily upscale old existing backgrounds:

Spoiler

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

But here are 2 problems:
1. Some details are lost because I didn't describe them.
2. Those old images are not perfect to begin with, so it might be better to create something completely different using some other images as a base.

So bc of this I need a detaled description of each background: what exactly should be there (all small details) and how the composition should look like (need a reference image). We can start with 2-3 backgrounds and see how it will go.

The main concern here is lore, bc I played it a long time ago and don't really remember anything. So we need someone who knows the lore to make those descriptions.

I should clarify for bystanders who read me that I find most of the current bgs to look satisfying to me, in terms of art style, composition, atmosphere and lore. Nothing should be replaced in the base-game unless it's unarguably better. But as a mod, it would be amazing to have a homogeneous art overhaul.

So sure, let me try to help. Do you want me to try here on the thread, or we do a private message thread to keep it contained?

Wherever, tell me the 2-3 backgrounds you want a description of. While we can use the current ones as models, we can also go from what I had in my mind or put as tags in my internet searches when looking for them, and see what fits best.

 

[bews]

Wherever, tell me the 2-3 backgrounds you want a description of. While we can use the current ones as models, we can also go from what I had in my mind or put as tags in my internet searches when looking for them, and see what fits best.

Just pick any that don't have nudity (since it's not supported yet). Also no need to for tags, just a simple text description. The more detailed it will be, the better.

 

[qwertyu12359]

Here's current Serpi: (cool because it's overcrowded and mixes architecture styles which infers it's fantasy-byzantine side of Eternal Rome) - I think it should be the composition reference.

Here's the original: (cool because we can see the anthil as per the lore descriptions, and the technology is visible too)

This is the opportunity to get the best of what AI can offer: an artistically superior and more lore accurate version than what we have available (without an artist team at hand).

I fed ChatGPT with as much lore as I was able to, and the two images, asking for a detailed but straightforward description of what we should be able to see at eye-level when entering the quarter:

Spoiler

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

Tell me if that's okay.

 

[Flying Geaser]

I hope to get a 3d or animated version with LLMs in the next 20 years.

 

[bews]

Tell me if that's okay.

That's not what I meant, guess I should have given an example. So if we take this picture

Spoiler

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

I had to put in the description that there is a staircase or it would morph to something unknown. Also I had to describe those trees as cypresses trees - otherwise they would become dead trees or shadows. This is the type of the description I need: what and where.

This is aboslutely useless and makes it worse:
" the air thick with the scent of oil and smoke "
"The entrance to the Serpentine Quarter" - AI doesnt know what is "Serpentine Quarter"
" The image also captures various smaller details " - this exact text makes it worse.

It should look something like this: An old worn building on the left with a big neon sign saying "Virgin's Hip". Two men in a futuristic armor are standing at the entrance. They have glowing red eyes and hold an energy spear with the glowing tip.
And so on.

Anyway I've still generated some images with the current prompt:

Spoiler

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

Some details were lost bc of bad prompt + the whole quality is not great bc a second (and maybe even 3d) pass needed. Basically the whole idea is turn the original image into something that AI made and then improve on it.

Or... We can generate from 0:

Spoiler

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

Results are not great bc the prompt is filled with the details that AI doesn't understand - this is why it can't follow the description properly. The longer the description - the less weight each detail has. Add something that AI can't understand how to draw and u get the results above.

Even so... IMO all of them are better than the original.

 

[qwertyu12359]

That's not what I meant, guess I should have given an example. So if we take this picture

Spoiler

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

I had to put in the description that there is a staircase or it would morph to something unknown. Also I had to describe those trees as cypresses trees - otherwise they would become dead trees or shadows. This is the type of the description I need: what and where.

This is aboslutely useless and makes it worse:
" the air thick with the scent of oil and smoke "
"The entrance to the Serpentine Quarter" - AI doesnt know what is "Serpentine Quarter"
" The image also captures various smaller details " - this exact text makes it worse.

It should look something like this: An old worn building on the left with a big neon sign saying "Virgin's Hip". Two men in a futuristic armor are standing at the entrance. They have glowing red eyes and hold an energy spear with the glowing tip.
And so on.

Anyway I've still generated some images with the current prompt:

Spoiler

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

Some details were lost bc of bad prompt + the whole quality is not great bc a second (and maybe even 3d) pass needed. Basically the whole idea is turn the original image into something that AI made and then improve on it.

Or... We can generate from 0:

Spoiler

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

Results are not great bc the prompt is filled with the details that AI doesn't understand - this is why it can't follow the description properly. The longer the description - the less weight each detail has. Add something that AI can't understand how to draw and u get the results above.

Even so... IMO all of them are better than the original.

I don't think any of them is better than the original, I'd like to know why you find so. The original (from 2.3 at least) is a good looking piece of art, with great composition and atmosphere. Looks like an oil painting, I feel like I'm there, feeling the crowd, lost in this byzantine pastiche from nowhere.

But I think they are close to be on at least on par. It has for itself that it can include technological elements and known landmarks, so the potential is definitely there to look better and ultimately replace it even in the official release.

The IA generated pictures for now mostly misses population density, guards look uneven with the serpis guards (which by the way was made with Midjourney by Uncelius with every detail matching what we wanted so it can be our reference for armor and colorimetry). Text can look weird, etc. I also like the water element of the original, which makes it fit the overhaul city map, and give this "crossroad of civilisation" aspect to it like Constantinople.

Once every element and the composition we need is there, it'll be a question of refining style, lightning, texture and atmosphere.

Here's a new attempt:

Spoiler

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

 

[OmarMammadov]

the pictures are not loading in dev version for me. how can i fix it? (i download everything)

 

[qwertyu12359]

the pictures are not loading in dev version for me. how can i fix it? (i download everything)

Please tell me where you unpacked the media_base archive

And do you have the "content" folder (inside "game" folder) next to the "css" and "engine" folders? Screenshot of your "game" folder is appreciated.

 

[joeshmo828282]

That 4th picture is scary impressive, even if the 'techno-sphere' is a bit on the nose.

I am wondering if you could run them though an oil painting filter to get them back to the oil painting style...

Spoiler

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

ehh. maybe? maybe there are AI oil painting tools that would do better? I suspect we might have to do some AI in-painting or perhaps actual artistry to touch these things up if we wanted to use them...

For example, if you wanted say, a top down of the entire city, I suspect you would have to do individual pictures of the fogs, slums, the wall, the quarters, and white town, and then merge them together. And the quarters all need their magical spire whatevers which hold back the fog, and white town the hippodrome and so forth.

 

[bews]

Looks like an oil painting

run them though an oil painting filter to get them back to the oil painting style

U guys don't get it: I intentionally tried to make it not look like an oil painting. Ofc AI can do oil style natively (even old SD1.5 can), but I intentionally pushed for anime realism so the new girls images would fit well and be in the same style.

Imagine this: u see a super cute girl in anime realism style and some background behind her. In this scenario u want her to fit into background, so the whole composition would look like a seamless picture. Bc of this oil painting doesn't fit in the first place - we want a background that won't look weird if we put a girl image on top.

I don't think any of them is better than the original, I'd like to know why you find so.

1. Similar anime realism style as the new girls images
2. Better quality (the original is so low quality that image recognition software can't even tell what is on the picture)
3. More details, including game lore text

Text can look weird

Text is not a problem anymore with the new model - even if it will look bad I can simply re-render that exact part and make it perfect. Same with other parts, but obv it will take more time that simply taking picture as is. Remember - this is not a proprietary locked model like midjourney, here we can edit even the smallest details in the image. So basically the workflow will look like this: first generate overall composition and then edit small details while keeping the image mostly the same.

I wil try new prompt later. Also if you need fog then add it to the description - new model can make fog no problem.

 

[OmarMammadov]

Please tell me where you unpacked the media_base archive

And do you have the "content" folder (inside "game" folder) next to the "css" and "engine" folders? Screenshot of your "game" folder is appreciated.

thank you for your reply but i fix it. it was download error. some files not downloaded when i first download it.

Silly me

 

[qwertyu12359]

U guys don't get it: I intentionally tried to make it not look like an oil painting. Ofc AI can do oil style natively (even old SD1.5 can), but I intentionally pushed for anime realism so the new girls images would fit well and be in the same style.

Imagine this: u see a super cute girl in anime realism style and some background behind her. In this scenario u want her to fit into background, so the whole composition would look like a seamless picture. Bc of this oil painting doesn't fit in the first place - we want a background that won't look weird if we put a girl image on top.

I do get it, that it's your intent to make it look like you did, as opposed to how it was. I didn't doubt.

What I'm saying is the original had its strength, and the style conveyed an atmosphere that is completely lost in the AI version.

While I'm not opposed to a change of style, it's good to know the pros and cons.

I also think that, within all the variations of the "realistic anime realism" style, the IA render is lackluster. I've seen better, in that style. It looks like it's from a 2006 VN, a bit blocky, and awkward.

Don't believe I'm shitting on your work and the renders, right now I'm talking in a position of a regular user with a subjective opinion, not as a developer vetoing anything. I invite you to continue in your direction even if that means "realistic anime style" or analogue, I'm thinking you can go even further in that direction to avoid a "worst of both worlds" phenomenon. Maybe we need other references pictures.

2. Better quality (the original is so low quality that image recognition software can't even tell what is on the picture)

Is it, though? Here is the

You must be registered to see the links

. Agreed it's a JPG, so probably a sample of the original high resolution image, still it's well done and professional. I just cropped it and saved it with lossless compression, so we didn't lose detail (mind everyone reading that f95 previews are often lossy and it's needed to open the image in a new tab).

What you are maybe saying is lack of quality might also be part of the style.

Text is not a problem anymore with the new model - even if it will look bad I can simply re-render that exact part and make it perfect. Same with other parts, but obv it will take more time that simply taking picture as is. Remember - this is not a proprietary locked model like midjourney, here we can edit even the smallest details in the image. So basically the workflow will look like this: first generate overall composition and then edit small details while keeping the image mostly the same.

I wil try new prompt later. Also if you need fog then add it to the description - new model can make fog no problem.

Cool!

 

[bews]

conveyed an atmosphere that is completely lost in the AI version.

Atmosphere is made by details: for example weather (including fog), time of the day, lighting, ppl mood, depicted actions, music, etc. It should be possible to recreate it in any style. If the generated images don't have it - it means we didn't specify the details clearly enough and didn't put emphasis on them. And probably our prompt is too long, so weight of each detail becomes less and some are lost. The way I see the correct way of doing this is first describe the scene properly and then add details as another pass. For example, if will ez for AI to draw any 2 men at the entrance. Then later we can change them to those we need, but their position in the image will stay the same. Good old "divide and conquer".

I've seen better, in that style. It looks like it's from a 2006 VN, a bit blocky, and awkward.

Could be true, since this is just a base model with nothing else. It was trained to be able to create everything, so obv it can't too good in anything. Tbh it's really impressive that the base model can even make this level of images out of the box. But I'm sure more specialized checkpoints will made later by the community, so the quality should improve too.

Why do it now then? Bc generation techniques will be the same later - only quality will be a bit better. For example if our prompt can't give us what we need - it won't be able to do it even with the specialized model. And the second reason is that I simply want to see what this new model is capable of and how it works.

till it's well done and professional.

Yeah, I argee, it's well done and professional... for a painting. I don't want a painting, so detalization is not good enough. Just look at these parts:

Spoiler

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

It's a blurry mess. It's so bad that AI can't even understand what is happening there.

I'm 99% sure that the original author started using paintings not bc he wanted, but bc there wasn't any anime alternative in the same setting. But now we can make the whole game have the same art style, just like any other professional game.

 

[qwertyu12359]

Atmosphere is made by details: for example weather (including fog), time of the day, lighting, ppl mood, depicted actions, music, etc. It should be possible to recreate it in any style.

I partially agree, atmosphere is enhanceable and a good one can be achieved in many styles...

But if you were to reproduce this famous impressionist oil painting: with photorealism or anime style... setting the same atmosphere, even by controlling the variables you mentioned, would be a horizon more than an attainable goal. Though,

You must be registered to see the links

when IA turns it into photos: -- Not that bad!

I'm 99% sure that the original author started using paintings not bc he wanted, but bc there wasn't any anime alternative in the same setting.

If you mean the author of JONT, he was indeed using a bit of everything for the background. Sometimes japanese VN styles , others are western 2DCG . It's one of the many weird quirks of this game that are endearing, and accidentally fit the "chaotic chaos of the multiverse's crossroad" which is the setting for Eternal Rome. No matter the backgrounds, NPCs like this or this never seemed to stick out from the environments to me.

Then for v2.0 I was tasked by Powerairmax to do a complete overhaul of every single background. I wasn't set on a style in particular, but western 2DCG was indeed more available and their concepts looked the most like what the lore was demanding. But I didn't "settle" for western painting compositions, I simply took what we found was best.

About blurry details, it's actually common to every backgrounds and many 2DCG on the internet. In real life, the background can appear blurry too, and so do moving faces in a crowd (if you recognize the same individual's face from the crowd every time you go to a quarter, it's uncanny, as a player; makes the image appear still and dull). That's quite important to the atmosphere. Though I get it's hard to work with if the IA can't make up what it is, still, it's not a lack of quality but an artistic intent, which happens to work well with backgrounds of JONT you end up visiting so often.

Nonetheless, I'm just giving a piece of my mind here, maybe it'll be helpful, maybe not. You retain full artistic autonomy over what you want to do.

Edit: the point I was trying to mean might not be clear. In short: I find your goal commendable (homogenization of art style across the whole game). But you sometimes seem to believe that the content present in the game currently (or before) was selected out of bad tastes or lack of means; I feel the need to try to make you aware of the strengths of what is currently in the game, so that you'll be able to offer strong replacements that tick every boxes of what the previous content conveyed, without the artistic shortcomings which are an inherent threat to remakes/overhauls.

 

[the_sexiton]

I'm looking forward to seeing the image pack. I think the largest benefit will be consistency between various training acts where it's clearly the same woman rather than using random pictures of different women for each bit of training. In terms of backdrop art, I think what is currently in the game is fine and maybe minor focus on higher resolution upscales or maybe a little sharpening could help them out. In order for new backdrops to work they would need to capture the rough and dirty look of the original backdrops. The original art looks much more gritty vs the clean look of the AI art. Even the white town has a rough and gritty look to the art that makes the viewer feel the grit of the stone without actually seeing it and despite that part of town being more clean and high end. This should look like a dirty and worn outpost of humanity that has regressed from current society in many ways. It's almost medieval in it's style and cleanliness. Hope that makes sense.

 

[bews]

New prompt is basically the same as the old one, so result didn't change much:

Here is the comparison with the same seed:

Spoiler

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

But still I generated some more images and some even look really nice and could be used as a base:
img2img using old as a base:

Spoiler

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

txt2img

Spoiler

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

 

[SilmeToO]

okay, some question
-How to not break sanity
-Fastest way to increase obedience

 

[qwertyu12359]

Little news: this is supposed to be the day I release Jont dev' version as is. However, Imperator told me he should have time this weekend to finalize the release. So I postpone to the beginning of next week.

------------------------------------------

New prompt is basically the same as the old one, so result didn't change much:

Here is the comparison with the same seed:

Spoiler

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

But still I generated some more images and some even look really nice and could be used as a base:
img2img using old as a base:

Spoiler

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

txt2img

Spoiler

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

Though there are some that are closer to my original idea, I think the one with the most amount of potential for "wow, looks good" is this one:

What it needs on top of that: 1) more technology (neons and stuff). Maybe localized to the right. So rather than small buildings, it would be the technodrome, with one entry to a huge indoor place. It would look sci-fi but still integrates with the rest.
2) A huge anthil building, probably on the left, visible in the background, that reaches the sky. Ideas:

3) Even more density of people, it should be swarming, disorganized, like the source picture. Boats on the river, people on bridges... maybe even some activity from the windows in the towers

4) What I identify as the serpi president tower in the middle of the background, should look more eastern (rounded domes for roofs)
5) On the right, maybe more building and less vegetation, or if vegetation, it should be part of a rocky mountain thing (on the map there's a rocky mountain parallel to the ground, I like the idea that white town is standing above all over cities, and you can see this giant mountain from below in some of the quarters: ).
6) One bird fused with the cable line from the building

 

[bews]

I think the one with the most amount of potential for "wow, looks good"

Yeah, this and the last one are my favorites too.

What it needs on top of that:

Give me the prompt for each part.