[mod] Stable Elves

[buttjuicejoe]

20 hours ago, lancarta said:

Hi! You did a really great job at this, and I am having a new motivation towards playing stellaris!

 

Im trying to make myself something like this, but with human ears and a bit more diversity... but my only computer is an average laptop, so I started paying for NovelAI generation. However, I was not able to achieve the same results you have given... may I ask a few questions and maybe a plea for help?

 

1. The keyword setup.

I was able to get see-through, or other 'simple' forms of clothing, but I was not able to get the intricate and beautiful dresses your characters wear, nor the various poses or facial expressions. I'm very new to this, and I would appreciate it if you could share your secrets on the keyword setting.

 

2. The possibility of acheiving decent quality under 50 steps.

I paid for the highest tier on NovelAI, so I get unlimited generation for images  512x768 and 28 steps or below. I read somewhere that StableDiffusion had options for way more steps and other settings, and I was wondering if something even close to your work is possible with the limitations imposed by NovelAI.

 

3. Background transparency.

How did you make your .dds files? Im quite new to image processing yet as well, and Id like any tips or suggestions you would give to a beginner for making the backgrounds transparent.

 

In short, I was wondering if I could achieve something even remotely close to your work with a paid NovelAI subscription... My idea is to create something similar with more images and more human-looking. I really would love to get your help on my new project!

 

Thanks a lot for reading such a long request!

not sure what novel ai has going on but i use stable diffusion, newer models and sampling methods work fine less than 50 steps, i usually run in 30-40 with dpm++sde karras sampler, not sure if novel has that, but stable has a variety of dpm samplers that all do good work in lower steps

for making the dds with transparent backgrounds i use paint.net, its a free software, basically one step above microsoft paint, but you can cut out images and do transparency and stuff

as for outfitts keep trying new prompts, for stable you can add loras as an extension to the program but novel prolly cant do that so just try creative prompting i guess. i like "intricate filigree lace", that ends up with neat stuff

[Armchair]

@lancarta

 

My images were generated locally using a copy of NAI's image model that was leaked last fall.  At the time that was arguably the "best" model for making anime-style images.  Since then I think NAI and other parties have created improved models, but for the sake of this mod I kept using the same model for the sake of consistency.  I know that people have managed to improve the quality of their outputs dramatically since I started, but I kept things the same for consistency and I haven't had time to learn the new stuff yet.

 

To answer your questions:

1. When picking your keywords for your prompt, try looking at Danbooru for examples of the kind of thing that you're looking for.  Then try using the Danbooru tags as part of your prompt.  NAI and other models often were trained on images that were taken from Danbooru and in turn the models end up recognizing the tags.  Most of the intricate details were generated as part of the upscaling process.  My images were originally generated at 448x640 and then were upscaled to 896x1280.  There's no significance to those precise values.  It was just the result of some trial and error to see how high that I could push things without experiencing crashes.  I'll post some example prompt settings from one of each leader type below.
2. 28 steps is honestly plenty.  You see diminishing returns on any changes really fast.  I think I used 40 steps on many of my prompts, but that was a number that I picked when I didn't know any better and I decided to keep it throughout the whole project since I figured that consistency of mistakes was better than doing things correctly halfway through.  I don't have as much experience with NAI image generation interface.  Some of the scripting options in the local interfaces are certainly convenient.  For example using X/Y/Z plots to queue up large batches of images with slightly different permutations of different settings so that you could compare results side by side and pick your favorites.  X/Y/Z was also useful for creating variety.  For example if you have a prompt that you like, then you might generate batches with slightly different settings such as swapping out "small breasts" for "large breasts" or "petite" for "tall" to try to get some variety of different body types that are wearing a similar uniform.  Or maybe some you run variants of a prompt replacing "topless" with "underboob" or "cleavage cutout", "impossible clothing" ,etc.  I used that script to generate the composite image below to show the difference in one of my empress prompts at different steps between 5 and 40.  As you can see, the final product gets 99% dialed in very quickly.  Past ~20 or so steps,  I don't think that any perceived improvements in quality can be directly attributed as being an increase to the step count and instead are simply a matter of the thousand monkeys getting lucky at that particular step count.   That said, if computation time is of no issue, regenerating an image at slightly different step counts can be a lazy way to try to fix an image that you mostly like but is "off" in some small detail.  For example, 20 steps has a bit of a problem with duplicate nipples and her nipples are clipping through her dress.  Rolling the dice again might fix that through random chance.  I don't think that a shortage of steps will be a bottleneck for you.
3. Background transparency was mostly accomplished via manual photo editing in GIMP.   I made my job a bit easier by using the term "simple background" as part of the prompt so that my images would have simple backgrounds that were easier to select with things like the fuzzy select tool. Eventually I had some luck with an AI-driven plugin for GIMP for removing backgrounds that worked ~20% of the time.  I've been told that some plugins for stable diffusion for removing backgrounds have been made, but I don't have experience with them.   Someone else recently asked me this question, so I'm just going to copy-paste my response below.

 

 

 

Below is my workflow for removing backgrounds (I sure hope that spoiler tags work on this forum)  (Edit:  I apparently botched the nesting of the spoiler tags, but I think the response is still legible for anyone who wants to read it.  I'm not seeing how to actually fix the spoiler tags here in the edit post options)

 

 

As for the manual steps in gimp, I mostly followed this workflow.

 

1. add an alpha channel to the layer if it isn't there already so that you can have transparency.
2. duplicate the original layer so that I have a master copy to recover from if I accidentally erase something that I don't want to.
3. between those two layers, create another layer of a solid color.  Pick something easy on the eyes that also highly contrasts with what's in your image.  I mostly used a shade of green, but it depends on what your image is.   This step is just so that you can more easily see what you've erased without straining your eyes too much.
4. Repeatedly use the fuzzy select tool to try to select big chunks of the background and delete them.   I had to play around with the threshold settings of the select tool to try to coax it into selecting what I want.  I mostly only care about the bits of background that is directly touching the subject.  This step mostly works because my backgrounds were "simple backgrounds" that were largely just simple color gradients or fields of solid color.  It wouldn't work as well on a complex background.
5. once I felt that I had removed most of the background, then I'd use the fuzzy select tool again to select my subject with the threshold dialed all of the way up.   I'd then copy her to a new layer and then hide the old layer.  Personally I kept a bunch of old layers as a sort of "checkpoint" that I could go back to incase I messed up.   Anyway, this step gets rid of any of the extra "trash" that was part of the background that wasn't caught by the previous selections since the select tool would only target whatever was directly touching your subject.
6. go back in with the eraser tool set to various sizes and trace touchup the details of the subject.
7. if at any point you realize that you accidentally erased something that you didn't mean to, use the lasso select to draw around your mistake, select the original source layer, copy, select your working layer where the mistake is present, and paste.  This will restore that chunk of the image and then you can try to carefully erase by hand again.
8. Once I was satisfied with removing the background, then I would feather the edges of the image so that you don't see a hard contrasting outline when they're in the game.  First I would duplicate my layer so that I have a backup to fall back to if I change my mind about something.  Then I use the fuzzy select tool with maximized threshold to select the character.  Then I go to select>feather and feather the selection by a few pixels.  Then invert the selection with select>invert.  Then delete.  You'll notice that the edges of the character will now gradually fade to transparency.  Play around with the number of pixels feathered to find a value that looks good to your eye.

 

Another note, when downsizing the portrait for use in the DDS file, there are different interpolation settings that determine who pixel color data is discarded as resolution shrinks.  Experiment with those to see which you like the most.  IMO I felt that "NoHalo" looked best and kept the final product from looking too fuzzy.

 

 

 

There might very well be much more efficient ways to approach this problem.  The above is just what I did when I was making stable elves.
 

 

Empress prompt

 

parameters

masterpiece, highest quality, solo focus, (skimpy:1.04), (short:1.15) and (toned), (chibi:1.04), (dark elf:1.08) empress with (medium breasts:1.1), high angle, (no panties), and (revealing clothes:1.1), (underboob:1.2), ( transparent silk pelvic curtain:1.3), (see-through:1.2), (pussy:1.2), (nipple slip), detached sleeves, (perfect face:1.1) intricate (high detail:1.1) body, thigh highs, (evil smile:1.1), (pubic tattoo:1.1), covered nipples, eyes up, lips, simple background nsfw
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 40, Sampler: Euler, CFG scale: 10, Seed: 1168125563, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.5, ENSD: 31337, First pass size: 448x640
 

 

Envoy prompt

 

parameters

masterpiece, highest quality, solo focus, (skimpy:1.04), (short:1.15) and (toned), (chibi:1.04), (dark elf:1.08) slut with (medium breasts:1.1), high angle, (no panties), (scandalous intricate cocktail dress),(revealing clothes:1.1), (pussy:1.3), (pussy peek:1.0), (nipple slip), (perfect face:1.1) intricate (high detail:1.1) body, covered nipples, clitoris, assertive female, eyes up, lips, (pelvic tatto:1.0), simple background nsfw
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 40, Sampler: Euler, CFG scale: 10, Seed: 468057538, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.55, ENSD: 31337, First pass size: 448x640
 

 

Governor prompt

 

parameters

masterpiece, highest quality, solo focus, cowboy shot, (skimpy:1.04), (short:1.15) and (toned), (chibi:1.04), (dark elf:1.08) matron with (medium breasts:1.1), (high detail:1.1) body, wearing an off-shoulder evening gown, side-slit, see-through skirt, (no_panties:1.1), (perfect face:1.1) intricate , eyes up, lips, (embroidered womb_tattoo:1.1), simple background erect_nipples_poking_through_clothes, assertive_female
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 40, Sampler: Euler, CFG scale: 10, Seed: 2512450143, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.6, ENSD: 31337, First pass size: 448x640
 

 

Military prompt

 

parameters

masterpiece, highest quality, cowboy_shot high_angle (skimpy:1.04), (short:1.15) and (toned), (chibi:1.04), (dark elf:1.08) slut with (medium breasts:1.1), ornate_intricate_red_crimson_high_collar_military_dress_uniform_jacket (impossible_clothes:1.2) see_through_transparent_dark_slate_bodysuit_plugsuit covered_nipples_poking_through_clothing lowleg_microskirt (pussy_peek:1.1) (womb_tattoo:1.1) (no_panties:1.1) thigh_highs , seductive_smile assertive female, eyes up, lips, simple background gold_trim
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 32, Sampler: Euler, CFG scale: 10, Seed: 619152548, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.6, ENSD: 31337, First pass size: 448x640
 

 

Civilian prompt

 

parameters

masterpiece, highest quality, solo focus, cowboy shot, skimpy, (short:1.15) (chibi:1.04), (dark elf:1.08) submissive slut slave with (medium breasts:1.1), seductive_smile pussy_peek cleft_of_venus clitoris nude high angle, (perfect face:1.1), (high detail:1.1) body, eyes up, lips, simple background, perky nipple, womb_tattoo, simple background, thigh highs choker lowleg microskirt detached_sleeves
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 36, Sampler: Euler, CFG scale: 9, Seed: 812070242, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.65, ENSD: 31337, First pass size: 448x640
 

 

Scientist prompt

 

parameters

masterpiece, highest quality, solo focus, cowboy shot, skimpy, (short:1.15) and (toned), (chibi:1.04), (dark elf:1.08) slut scientist with (small breasts:1.1), high angle, (labcoat:1.2), (turtleneck_sweater:1.3), (cameltoe:1.2), sexy_glasses, (perfect face:1.1) intricate (high detail:1.1) body, eyes up, lips, simple background, (perky_nipple_slip:1.1), clitoris, lowleg navel cutout, womb_tattoo
Negative prompt: multiple girls, lowres, text, error, missing fingers, (bad anatomy), extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, extra fingers, mutated hands, (multiple penises), (disembodied), (poorly drawn hands), (poorly drawn face), (mutation), (deformed breasts), (ugly), blurry, (bad proportions), (extra limbs), blowjob, deepthroat, animal ears, extra ears, panties
Steps: 36, Sampler: Euler, CFG scale: 9, Seed: 385401564, Size: 896x1280, Model hash: 925997e9, Denoising strength: 0.55, ENSD: 31337, First pass size: 448x640
 

Edited May 15, 2023 by Armchair

[Armchair]

I suppose I have some more details that I want to add after thinking for a bit.

 

1:  I suspect that you can most likely create outputs that are similar if not better in quality than what I created for this mod via NAI's services.

2:  My settings weren't some kind of "secret sauce" to get good results.  I didn't get gold every time that I hit the generate button.  My 1080 TI spent several days generating images for this project.  I have thousands of elf images eating up several gigabtyes of space in my output folder.  The images within this mod were what I considered to be the cream of the crop and the overwhelming majority of outputs were flawed in my opinion.  Good settings can increase the likelihood of getting a good output, but it's still very much a gacha game.  So don't be disheartened if any individual prompt is flawed.  Most of mine weren't good either.  When you think that a prompt is producing outputs that are "close to being good", then keep rolling the dice for a while.

3:  I don't know what options that NAI has, if any, regarding "image to image" generation or "inpainting" options.   IMG2IMG and inpainting provide useful tools for touching outputs that are mostly good but have a few flaws.   That said, IMG2IMG and inpainting aren't necessary to create a mod like this.  Only a small handful of the portraits were touched up, and 90% of the touchups were fixing situations where the top of somebody's head was out of frame.

Edited May 15, 2023 by Armchair

[XavierMace]

My 2c here, I can also only speak to Stable Diffusion.  Regarding the number of steps, there's a variety of factors.  Sometimes 20 will be fine, other times you're questioning if 50 is enough. As Armchair said, you're not going to get gold every time unless you're really unspecific about what you're looking for.  IE just "naked woman" will get you pretty reliable results (as far as lack of defects goes).  I strongly recommend going over to https://civitai.com/ to find some results you like and see what settings they used. 

 

But once you start tacking on prompts and/or LoRA's (body/breast sizes, ethnicity, age, etc) it starts struggling more to give you exactly want you want.  I do think anime/cartoon style images generally have less issues with fine body details, especially if you're using an adult oriented checkpoint.  But, sometimes you'll get defects that are so random you almost want to keep them.  I've been considering doing a "mutants" portrait set using assorted images with defects.  IE, I've got got one with a woman with a tiny leg where her clit should be.  I've also got a woman sitting in a chair where the chair has straight up human feet.  I've got a woman who has a second torso where her neck should be complete with head and arms.  I've got women with their butts on the front of their bodies.

 

Right now for my mod I'm going with ~50 portraits per leader type and I've got a few revisions and types.  I generally create a 100 portraits at once knowing I'm going to be throwing a bunch out.  I'm currently at 3,900 images (3.8GB) and I've actively got generation running in the background:

 

I'm still tweaking my process but I do IT for a living and I'm impatient, so manually editing images wasn't appealing to me.   I have no doubt that Armchair's manual results have the better image quality, so use at your own discretion, but my whole process is automated.  I can't take credit for figuring any of this out on my own, but I'm a master Googler.  Basically I'm using a couple of utilities to do the editing and conversion and a PowerShell script to do the renaming.  I'm currently generating my portraits at 712x570 (1.5x expected size) as Stable Diffusion seems to have less issues creating defects at the higher size due to most models/checkpoints being done with 512x512 images.  Here's the basic flow:

 

Bulk CLI Remove Background (https://github.com/danielgatis/rembg/):
rembg p C:\examplepath\admiral C:\examplepath\edited\admiral

 

Powershell bulk rename:
ls | %{Rename-Item $_ -NewName ("SpeciesName_Sci{0}.png" -f $nr++)}

 

ImageMagick bulk conversion (https://imagemagick.org/script/mogrify.php)
mogrify -format dds *.png
mogrify -resize 475x380 *.dds

 

The primary caveat with this process is if your backgrounds on your image have really similar colors to your non-background parts, it may not be able to tell the difference and remove stuff you didn't want removed.  When that happens, I've been scrapping the whole batch and trying new prompts to get a background that doesn't cause that problem because that's still quicker (at least for me) than manually editing 100's of photos.

 

Edited May 18, 2023 by XavierMace

[Armchair]

I may need to play around with something like that, but I had to make use of a manual process at the end to make sure that my favorite images worked since my attempts with automated background removal had a low success rate and I wanted my favorites to make it in.

 

Part of this project was to give me something concrete to work with for playing around with Stable Diffusion.  Such as trying to touchup images with IMG-to-IMG inpainting.

 

Here's an example with one of my favorite images from the envoy set.  Left was the original, and right was the result after some inpainting to fix missing nipple  and clean up the crotch.

 

Going completely automated and only keeping the subset that successfully make it through the whole process is probably the way to go if you're just trying to churn out a lot of species, but that wasn't really my goal.  When I started making this mod, the mod wasn't the end-goal.  My goal was learning about SD, and the mod helped give structure to that learning process by giving an objective to work for and provide problems to solve.

 

Edited May 19, 2023 by Armchair

[XavierMace]

Yeah, like I said, there's no question your manual process is going to produce higher quality results.  I was just throwing that info out there in case the other curious people lack the time/interest in manual editing.  The automated background removal is pretty overzealous and I haven't even touched inpainting yet.  But given the low resolution of Stellaris portrait, I think I'm about to a point where the automated quality is "good enough" for me at least.  I definitely have a variety of other species I want to pump out first.

[Armchair]

While playing with the paragons DLC, I realized that one of the admirals has a bit of her face accidentally deleted by the eraser brush.  If I have time this weekend, I'm going to try to find the original source image and make a new DDS file for her.

 

 

On an unrelated note, I found out that apparently level 5 leaders can appear in the leader pool?!?!  I was looking over my leader pool when starting a new game, and I was startled to see a General with the Kidnapper trait available for hire on day 1.  She also cost damn near half of my day 1 unity output in upkeep costs.  I hired her near the start of year 5 before the pool refreshed.  Hoping that free pops from wars makes up for the unity upkeep slowing down my tradition trees.  I don't think any of my mods would affect the contents of the leader pool.  I don't think that my civics should either (distinguished admiralty and that new heroic storyteller one).

 

edit:

also, it looks like UI dynamic overhaul updated something and now most of my portraits are off-center when viewing who is assigned to lead a plane/fleet/science ship.   Unsure if I'll need to re-align everything and re-release.  I'm going to wait a bit and see if anything else changes before I start experimenting with changing the DDS files in bulk.

 

edit 2:

It just occurred to me that maybe distinguished admiralty might be why I saw the level 5 leader in the pool.  I'm thinking that maybe the leader pool might population things with stuff up to 2 levels above the starting level, so "distinguished admiralty" starting my admirals/generals at level 3 might be allowing a level 5 leader to be in the pool.

Edited May 26, 2023 by Armchair

[XXXXX_EH29RR]

is it updated to the latest version of stellaris?

[xwhitepearlx]

4 hours ago, XXXXX_EH29RR said:

is it updated to the latest version of stellaris?

it can be used just slightly broken because the portraits are off-center, read the note from the developer above your reply

[Armchair]

I know I said I was going to update that portrait back in May, but then I had to deal with a Karen from the HOA and ultimately Stellaris dropped to the bottom of my priorities list.

 

I have not yet done anything about the off-center portraits.  When I last was keeping up with things, the game was in a bit of a flux between paradox making additional changes to the game and UI Overhaul Dynamic was also undergoing changes.   I suspect at some point in the future I might resize/re-center the images since them being slightly off-center bugs me a bit too, but I just have more pressing projects IRL to work on.

 

The update today only adjusted one of the admiral pictures.  I reduced the feathering on her since the old image made it look like there was a hole in her face.  I guess her chin looks a bit weird still, but this character's chin is narrow and it looks odd when scaled down.

 

 

 

 

On another note, I'm contemplating making another species.  I want to make a custom set of portraits to be used for my prepatent species when I play as necrophages.  Since I normally like using ocean homeworlds, I'm leaning towards trying to make a set of splatoon style inkling squid-girls.  This might include creating slavery portraits in addition to the normal archetypes.  I'm thinking their rulers could have a socialite theme, maybe their scientists could have a marine biology/wetsuit theme, and perhaps the military staff could look like they were in a watergun fight with wet-shirts/ink splotches/super-soakers.  Please don't hold your breath though.  Those are just some ideas bouncing around my head, and like I said I've had more pressing projects taking priority IRL.   I did do a fresh install of my stable diffusion setup the other day and started playing around with loras and experiment with different image models.  During that process I learned that my VAE wasn't properly configured in my original stable diffusion install (!!!).

[Calcifire]

12 hours ago, Armchair said:

 

On another note, I'm contemplating making another species.  I want to make a custom set of portraits to be used for my prepatent species when I play as necrophages.  Since I normally like using ocean homeworlds, I'm leaning towards trying to make a set of splatoon style inkling squid-girls.  This might include creating slavery portraits in addition to the normal archetypes.  I'm thinking their rulers could have a socialite theme, maybe their scientists could have a marine biology/wetsuit theme, and perhaps the military staff could look like they were in a watergun fight with wet-shirts/ink splotches/super-soakers.  Please don't hold your breath though.

 

I hope you do. The ai generated mods have been real pump and dumps. I've been hoping you'd be the one to actually stick around and make more content for the good folks round here.

[Armchair]

8 hours ago, Calcifire said:

 

I hope you do. The ai generated mods have been real pump and dumps. I've been hoping you'd be the one to actually stick around and make more content for the good folks round here.

Thanks.  That's a big compliment.  I did notice that it was kind of like the floodgates opened shortly after I released mine.  Not all of them have really been to my tastes, but everyone has their own preferences.

 

I've started trying to make some squid girls, but that batch didn't turn out very well.  I'll share some samples if I can produce anything that I like.  I just have a backlog of projects around the house to work on, and I know a lot of my videogame free time in the coming weeks are going to be consumed by War Thunder's summer event and the new Armored Core game.

[Armchair]

I've been toying around with the prompts a bit.  I'm posting some of the work in progress generations since it sounds like there is some interest.  I'm starting with the military class, and I think I'm going to just have a singular "military" set of portraits rather than distinct admirals/generals.  That's the way that I did things with the elves, and while I toyed with the idea of creating distinct generals, I don't think it will be worth the effort considering the magnitude of the role that generals fulfill within an average stellaris game.

 

As for the look itself, I originally was intending to stay true to source material and have the military characters look like they came out of a splatfest.  Basically wearing street/sport clothes, but splattered in ink.  But I've had a bit of an issue trying to get the characters sufficiently splattered to my liking and I'm not quite sure if I liked the way things were looking.   As an alternative idea, I decided to try to give them a more military uniform look, and by military uniform look I mean "Cammy from Street Fighter."  I think I'm liking the Cammy look better, but it needs some more tweaking.  I tried giving them camouflaged bodypaint, but it just started making their leotards body-paint "leotards".   I haven't decided if I want to try to discourage the "clothing is body paint" look yet, but it raises the idea of maybe trying to go for a full body war-paint look.

 

Once I get settled onto a look and find a prompt that can reliably create it, then I'll need to make variations of the prompt to mix things up a bit.  Things that will change the character's build, body language, the cut of their outfit, and their facial expression.  The goal is to maintain a consistent overall theme and look while also introducing enough variety to avoid what I think is the cause of the "pump and dump" feel that Calcifire was describing.   I think that a deliberate effort is needed to remix the prompts to produce variety, otherwise you fall into a trap of the characters looking a bit too similar and then at a glance they just start looking reminiscent of simple recolors.  I think that effect is actually outright visible below since each grid was produced by a singular prompt.

  

 

 

[zb123thecat]

On 7/25/2023 at 12:32 AM, Armchair said:

Once I get settled onto a look and find a prompt that can reliably create it, then I'll need to make variations of the prompt to mix things up a bit.  Things that will change the character's build, body language, the cut of their outfit, and their facial expression.  The goal is to maintain a consistent overall theme and look while also introducing enough variety to avoid what I think is the cause of the "pump and dump" feel that Calcifire was describing.   I think that a deliberate effort is needed to remix the prompts to produce variety, otherwise you fall into a trap of the characters looking a bit too similar and then at a glance they just start looking reminiscent of simple recolors.  I think that effect is actually outright visible below since each grid was produced by a singular prompt.

 

I think some consistency and limited variety is good. it makes each job and species more easily recognisable. So that someone is immediately identifiable as a scientist, governor or admiral, plus what species they belong to. 

[Armchair]

I'm afraid that my progress hasn't been anywhere near as fast as many would like since this has been a bit of a minor project of mine and I've had other things take precedence, but I wanted to share a few images.  Right now I'm trying to make some images of enslaved inklings that I personally want to use as a prepatent species when I play as Necrophages.  I've been experimenting with different models and loras to see what kind of results that I can get.   Spoiler:  I'm not actually satisfied with any results yet.  When messing around with AI art, you get a lot of garbage images.  But a particular combination of models and loras produced a few exceptionally silly results and I wanted to share them.

 

  

[Armchair]

Here's some examples of some better outputs.  I'm trying to go for a look for generic pops here.  But I'll need to mix up the poses/outfits/body styles a bit to get some variety.  Also tentacles/malformed biking equipment keep showing up in the outputs so I'll need to roll the dice enough to get a big enough pool without random props spoiling them.

 

 

 

I haven't settled on looks for scientists/military yet.  

 

For the rulers, rather than having a distinct uniform look I might decide to just go with a wide variety of unique styles since the splatoon idols tend to have unique looks.  Montu's clip the other day showing off how a scientist was governing a planet is giving me pause about giving any thought towards governor looks.  I might just have the rulers and governors share a pool of portraits?

 

I'm also not sure if I should rename this mod when I eventually implement these inklings, or if I should release them seperately?

[kingslime_nsfw]

Hi! I'm using this mod, and all of the leader portraits are the same as the ruler's? Any idea why this happened? Thank you!

 

[Armchair]

Off the top of my head I'm not sure what's going wrong for you.  I'm not sure if an update could have broken something.  I haven't been playing stellaris recently.  My youtube feed had a video from Montu Plays that talked about upcoming changes where they would be replacing governors with some kind of bureaucratic "official" class.  I imagine that I will need to update the mod in response to that, but I don't know when that patch is supposed to go live.  My video says that he was talking about stellaris 3.10.  Is that live yet?

 

Maybe it's not loading properly and is just defaulting to using whichever portrait that you picked for the leader?  The ruler portraits shouldn't even be eligible for usage as scientist/governor/military jobs as defined in the portraits.txt file.

 

Do you have any other mods enabled?  My go-to plan for troubleshooting mods is to make a smaller modlist where I just gradually add mods one by one as I make sure that nothing breaks with the introduction of each new mod.  I know some of the other AI portrait mods basically copied my files as a template, and for at least a while some of them didn't define new variable names for the definitions within the text files.   I think that they've fixed that by now but I haven't really been keeping up with everything.

 

Basically my advice is to just try to isolate if any mods are causing problems by just gradually adding mods to your load order one by one or in small batches.  I'm curious if this issue persists if you were to load up the game with just this mod?   Irony mod manager has some tools for checking mods for conflicts, but I don't feel comfortable describing how to use them.

[Armchair]

To be frank, I'm not sure what may or may not break with the upcoming 3.10 update.  Since they're consolidating some of the leader roles, I might have to end up merging the existing leader/governor/envoy portrait pools together.  I'll try to fix anything that breaks.

[Woitek]

Since 3.8 update Leaders, Council, Species select tab are offcentered and it's driving me crazy.
Such good mod borked by paradox portrait changes

[Armchair]

I'm going to try to do something about the off-center portraits this weekend.  I need to debug something that broke with my mod-list, and I've just been pre-occupied with obligations to family and work recently.

[Armchair]

So I'm not entirely sure what the "correct" dimensions are supposed to be for a portrait, so I tried just cropping some of the excess space to the left of the characters.  Loading up the game with no other mods, that seems to have fixed them being shoved off the side of the panel.   I tried scripting the process of cropping every portrait, but there are ~60 or so that I need to go back in and manually crop.

 

 

It also seems like, in the vanilla game, the tops of their heads are still being chopped off a bit.  However in most contexts their heads are fine when using UI overall dynamic, which is my preferred way of playing the game.  I haven't decided if I to fix the tops of their heads in vanilla or not since that would make them a bit smaller when using UI overhaul dynamic.

 

 

vanilla UI

 

With UI overhaul dynamic

 



 

 

 

[NinthMuse]

Hello. For some reason, I can't quite figure out. I can only see 21 of the portraits. Any help would be appreciated.

[Woitek]

9 hours ago, VoidStorm975 said:

Hello. For some reason, I can't quite figure out. I can only see 21 of the portraits. Any help would be appreciated.

Did you read mod description before download?
 

Quote

• 21 Rulers:  They have elaborate and ornate clothing, but the outfits are completely shameless in their use of silky sheer material.  Kind of a harem girl look.

 

[Armchair]

11 hours ago, VoidStorm975 said:

Hello. For some reason, I can't quite figure out. I can only see 21 of the portraits. Any help would be appreciated.

Woitek is correct.  As written, only those 21 portraits are defined as rulers.    Once you start the game, the other portrait sets will be used for different roles such as the miltary sets automatically being used for your Admirals, the Scientists for your Scientists, and so on.

 

If you wish to expand the list of portraits that are eligible for usage as your starting ruler, then edit the gfx\portraits\00_StableElf_portraits.txt file.   Look for the line "leader_class = ruler" line.   Just below that line is a line that says "portraitrs = {" and then there are several lines calling out which portraits can be used in this role ("StableElfRuler00, StableElfRuler01, etc").   Just extend this list with whicever portraits that you want, such as StableElfSci00 or StableElfMil01, etc.

 

I'm planning to release the updated mod later tonight where the portraits are re-centered.  I just need to kick the tires a bit first.