>>17770 I generated many more than this, but the quality goes downhill from here.

>>17763 I was going to censor it anyway for surprise, but forgot too. No, issues at all. >I wear glasses myself, and find girls with glasses attractive is all. :P

>>17763 That all promising, maybe for animated characters at some point. Virtual waifus, etc. However, it would be much more useful to have a system that would create 3D models of 2D faces, ideally in a file format that could be used to work with.

Did you use something like Lightning AI? What do you think of that? Seems to be easy: https://youtu.be/Xb7ucqIjjE4

>>17782 I used https://beta.dreamstudio.ai/ >They give you 300 free credits. <The following images are from openAIs Dall-E 2 with the prompt... >"an engineering blueprint showing the exploded view of parts needed to build a feminine android, lineart"

>>17910 >>17911 >>17912 >>17913 >>17914 Wonderful stuff SoaringMoon, thanks! This is just the type thing I meant when I said "his tech can be useful for much more than simply portraits of your waifu" (>>17784). I've already seen some absolutely world-class design work with this tech, and I'd encourage anons here explore it's boundaries. If it won't fit in here b/c NSFW materials, then post it over on the /monster/ robobutt threads.

>>17914 The sources of these drawing would make some nice wallpapers or screensavers. Anyways, if you're active in that area of research, consider getting us to a point where we can get a (simplified) 3D file from objects which were detected in a picture by something like Yolo, with additional information given, on the distance or the size of the object. If we get that, the next step would be adding additional infos and be able to add such an object to a similator. So the robowaifus could imagine to manipulate it, simulating it in their system, before they start doing it in real life.

A few more with refined key words.

>>18881 >A few more with refined key words. Thanks, looks really good and interesting.

>>18881 >>18882 >>18883 >>18884 >>18885 Naicu, thanks!

I think as the character Alita has shown us, as well as several other media pieces (including the works of the artist Sukabu) have shown us, that as long as the face is cute, then we have a high-tolerance for the rest of the body being quite machine-like for our robowaifus. > This may just be my own personal tastes biasing my viewpoint, however I suspect we are all pretty much the same in this regard. Comments?

>>21100 generally I think you're probably right, although I'm personally hoping for full on westworld waifus. once we have human-like faces (which is probably the hardest part considering the musculature and fine control required), I cant see the rest of the body being that much more difficult to skin. Cost might be one thing, depending on how cheap we can make near-realistic skin.

>>21107 >(which is probably the hardest part considering the musculature and fine control required), If professional 3D animation demands for hero facial work is anything to go by, then it is indeed the hardest part. By a long shot. Hands are next. I'm currently tackling understanding good hand designs. >Cost might be one thing, depending on how cheap we can make near-realistic skin. For this first set of prototypes, the skinning should be the absolute bare-minimum needed to protect the robowaifu's internals from direct damage, debris, dust & the like. Once we've solved most things sufficiently well to seriously consider tooling up for kits manufacture & distribution, then skinning will get more attention. R/n we have much, much bigger fish to fry. :^)

>>21109 >If professional 3D animation demands for hero facial work is anything to go by, then it is indeed the hardest part. I could see something like a facial mocap data set being used to train a neural network, same thing potentially with hands. The problem there being getting a mass amount of quality data. Or just decent recognition of faces from movie/TV and using that as the dataset for facial muscular movement. >For this first set of prototypes, the skinning should be the absolute bare-minimum needed to protect the robowaifu's internals Agreed, Panels are probably best for first few generations, other things like bipedal balance and natural gestures are gonna be much more important to create a decent simulacrum (atleast imo) I don't any experience with the current silicone used for higher end dolls (platinum cure i think), but I'm not sure how skin-like it really is. I wonder what the best product will eventually be (which will hopefully have embedded touch and temp sensors)

>>21112 >I could see something like a facial mocap data set being used to train a neural network, same thing potentially with hands. The problem there being getting a mass amount of quality data. MoCap data would indeed be a very natural data fit for ML IMO. When I'm looking into facial later I'll see if there are some good data sets available for us (I expect there will be). >Agreed, Panels are probably best for first few generations, other things like bipedal balance and natural gestures are gonna be much more important to create a decent simulacrum (atleast imo) Yes to both. In the meantime we should be able to use LARPfoam for Con-costuming, etc. It's acceptably lightweight, compliant, & cheap. >I wonder what the best product will eventually be Medical-grade prosthesis silicone is my bet. But again, that's for later on down the line.

Bing Copilot in creative mode: Draw me a picture that looks like s mural, and includes Cameron from Terminator SCC, Athena from Tomorrowland, Apple from Turbokid, Yumemi from Planetarian, Chii and Sumomo from Chobits, Alpha Hatsuseno, Mimi Takaoka from Button! CPU, Haizakura from Prima Doll, RyuZU from Clockwork Planet, and Lamia from Raised by Wolves Hmm, okay ... These are some wild hallucinations. I'll try again in a less creative mode. Also, it were four pics but I lost one, because of a reset and it seems to not have a history function.

>>22544 It's trash. And exactly because of their training to make things unbiased. I think I know what the did. They filtered out gender and race from every character, and only allow to ask for more "protected minority" representation. So either Dall-E or the Bing Copilot is basically like Disney.

Masterpieces...

>>22556 Don't worry, I won't go on spamming this thread with failures. I need to stop and do something else anyways. That said, some tips: It can lookup characters online, which doesn't help to much, but maybe if one would making it looking up the actresses as well? It also accepts anime-like and waifu-like faces as inputs, to somewhat avoid race switching towards black or adding males. It always wants to add faces looking like a robot, often like a helmet. No helmets or robot heads leads to you getting one face out of some imaginary mechanisms, but no waifu face.

>>22557 Thanks Noidodev. Honestly, I expect this trend to continue of gimping their AI systems. We predicted this stuff years ago. It's only going to get worse for everyone who relies on the Globohomo systems (all of the FAGMAN organizations are part of the GH). This is why we must have free & open systems. It's the only way forward for anon (indeed for the common man).

>>23743 I rather like that 2nd one.

>>23808 I really think we should go with the 'animu eyes' aesthetic for our robowaifu eyes, especially during these early years of prototypes/Model A's. Its a good engineering & design choice for a number of reasons IMO. Interesting that Disney, et al, got it right back at the very beginning of animation, right on up through to today since the baste Nipponese continued on with those traditions.

>>23808 That last one grows on me, actually. :^)

>>23810 Yeah, I have more good pictures like this, but I'll plan to use them for my comments. But I also started to collect some hilariously bad one, because some where just to funny to throw them away.

>>23811 >because some where just to funny to throw them away. When blithely jaunting along towards lofty goals of the sublime, during the pursuit of beauty (which we in effect are doing here), if one suddenly stumbles upon an open mockery of it, it is both jarring and a degeneracy to the soul. >tl;dr Heh, if you were /kind/ to all of us, you'd spoiler them if they're here purely for comedic effect. :^) >=== -poetry edit

>>23812 >you'd spoiler them if they're here purely for comedic effect. I thought you would think so. Okay, next time.

>>23808 I really have to stay away from this website. Spending too much time there... I made more than 180 pictures so far.

>>23833 I rather like the 3rd & 4th ones, with some adjustments they could be usable for a good robowaifu. >also Lol, the ones you're doing that look just like the faces of the marionette movie World Police Force. :^)

>>23835 >Rule #1, Anon Wow, those are really interesting-looking Anon. I relish the ones that show much of the technical 'underpinnings' of a good robowaifu. Aigis is our heritage canonical example in general, of course.

>>23836 Yeah, pretty good shit that I can do thanks to https://github.com/AUTOMATIC1111/stable-diffusion-webui

>>23837 Neat! Bookmarked under 'RW'. I neglected to mention the illustrious artist Sukabu, who does some tremendous robowaifu illustrations.

>>23838 >Neat! Bookmarked under 'RW'. You should also check out https://github.com/oobabooga/text-generation-webui for AI chat, it is damn slow with CPU, but if you have an nvidia card, it should be a breeze, amd in the other hand you need to jump some hops to get it working.

>>23839 Thanks! That's gotten better since the last time I looked at it. Nice job.

>>23835 Are these seriously generated? Wow. Soon you can start making a manga, but in color. >>23837 I know it's good, but I never planned to play around with generative AI for making pictures. Just slipped into it by accident.

>>23834 > that look just like the faces of the marionette movie Yeah, the first one? She's really cute. In that "Team America" movie the women don't really look that cute. You could also look at some pics from "Dark Crystal" where the females have very cute faces, though it's generally more for the /monster/ department. Also, "Star Wars: The Clone Wars" have such a look, though it's 3D animation. Generally I had the idea of mixing actresses when using this website, so they don't look too similar, but also the model there doesn't know how they look and really messes them up. Many imaginations of actresses look quite bad there, maybe they introduced a lot of noise on purpose. Kristen Bell and Blake Lively create absolute horror results, and "Sansa Stark" character has some filter making her look demonstratively old. It could also be that actresses often have some special feature or a mix of features which make them look good, but if you get those wrong, do something just a bit to much, it gets ugly pretty fast. These unique but beautiful looks are in a narrow valley so to speak. I gave up on using that model on the site to make face pictures that look like the actress by putting her name into it. Instead I just used that with some other terms and looked for how the output would generally look. Even just adding in the term "youthful" can help a lot. Generally some of these results can still be looking good and have certain features, even if it's not looking like the real actress. Adding in some name into a prompt still makes a difference in looks and then mixing them with "if XX and XX had a daughter together" also creates some nice results. I myself had the plan for a long time, to have a model that groups actresses and other good looking women together if they look similar or have some shared traits, then create new faces looking similar but not too much. So we can use those and also they won't look all the same. The third picture above (Scarlett inspired) also has some vibes of that marionette look. Aside from the charms this kind of look has, this could be a useful approach for a smaller waifu, since they look less like a anatomically correct human. It could also be the case that making the faces very realistic in terms of skin would be too difficult, and since the facial expressions would not be good enough in an early stage, it might put some people off.

>>23842 >It could also be the case that making the faces very realistic in terms of skin would be too difficult, and since the facial expressions would not be good enough in an early stage, it might put some people off. IMO we have ample evidence that's not the case. You can achieve wonderful animu appeal without coming anywhere close to the uncanny valley. Intentionally uglifying models just to prevent the possibility of it is quite misguided, I think. As to your other question in /meta, for my part, I've already said my piece: (>>23812) Good luck with your prompt explorations Anon! :^)

>>23855 >Intentionally uglifying models What I wrote was maybe not very clear. I just contemplated the level of human-likeness vs some more doll (or marionette) -like look which can also have it's charms. Including skin which looks really realistic vs a bit more doll like.

>>23867 Nice, but too robotic for my personal taste and the shoulders in picrel 2 are too wide.

>>23867 Those last two are excellent Anon. To my thinking that last one is arguably a better outcome for the treatment of anime eyes than even the highly-paid designers came up with for the feature film Alita. In that film, Alita still has a frog-eyed look in many of the shots. But the one you've provided here seems just perfect. Clearly very humanlike, without even a tinge of the grotesque to it. I don't know what model/prompts you're using for that beautiful character image, Anon, but please continue with it! Cheers. :^)

>>23857 I understand. And tbh a) my understanding wasn't likely very clear, and b) my own writing definitely wasn't very clear. I simply want to promote beauty here on this board. While I freely admit this is quite a subjective matter, however I'm fairly confident that the majority of men share my general tastes; regarding both a female's body and soul.

>>23874 >I don't know what model/prompts you're using for that beautiful character image Check the exif data on those pics know the prompts, models, seed, etc used. There is an extension that adds exif data on the images you make, so that everyone can mimic or derive from those pics. https://github.com/AlUlkesh/stable-diffusion-webui-images-browser

>>23887 Oh cool. Thanks I didn't know about that.

I only post the best failures if they're are also interesting and maybe funny. Spoilered then. Here in the first picrel I asked for a robot with female thighs and I got what I asked for... The other one was more of an happy accident, looks nearly flawless.

>>23902 That second one looks pretty good Anon.

This online image generator here https://huggingface.co/spaces/dvruette/fabric called Fabric works quite nice for portraits. Didn't test it much yet, for example not with many different variants of styles, anything else than portraits, or more realism. No time. Downside compared to open.ai is, that it doesn't make the prompt into the filename. Also it takes a while longer for each result. But the image quality is way better.

Are there any good prompt optimizers? I just really don't want to bother learning how to do it well when I'm sure that in a few years there won't really be a point to it anymore.

>>24335 >good prompt optimizers What do you mean? A software which helps you? There are subreddits which are about prompts, certainly other sites as well. >in a few years there won't really be a point to it anymore. These few years might be the important one, and the one where one could make money. Also, it might not be to hard. What do you want to do? Just portraits? That's not hard. My prompts are actually in the (open.AI) filenames. Just replace or remove the names of actresses or other elements. > Portrait of seductive and youthful Jailbait female, with realistic Alita doll eyes, light hair, and silicone skin > Youthful Scarlet Johanson, with realistic Alita doll eyes, light hair, wearing a dog collar > Digital art portrait of seductive and youthful Jailbait offspring of Summer Glau and Kristen Bell, with realistic Alita doll eyes, light hair, and silicone skin Fabric also takes examples which you can upload.

>>24335 >prompt optimizers? Just got this recommended: https://www.youtube.com/watch?v=nagWNxkVp54 Just look around in bigger forums. This place here is for sharing ideas, not the best place to ask and get the best advice on everything, since we're not so many people.

>>24338 >Also, it might not be to hard. What do you want to do? Just portraits? That's not hard. I struggle to even install stable diffusion without a tutorial, let alone actually use it right. I struggle with ChatGPT just trying articulate my thoughts in plain English to give it feedback. My problem is really that I don't just want *a* picture, but have an incredibly specific idea of what I want to see and have no way of getting Stable Diffusion to make something even approximating what I'm thinking of. (I'd probably even fail if I tried to commission something from a real illustrator) Although, a lot of what I'd like to make right now is more like SoaringMoon's engineering blueprint pics, or something more like concept art or reference images that could be for making 3D models. >>24340 I was thinking something more like https://youtu.be/bj9j3_lRL2Y that uses ChatGPT to help create prompts for Midjourney. And what I mean by "that in a few years there won't really be a point to it anymore" I mean that it probably won't be long until text-to-image models basically have LLMs integrated by default, so instead of using keywords separated by commas like a 90's search engine, any boomer can use them like the old lady at my job saying into her phone "Okay google, what was the name of that TV documentary from the 80's about the Kennedy assassination?", but instead it'll be me giving a 2-page long description for a character design reference sheet.

>>24341 I don't think I can help you with that. That said, this here: >more like concept art or reference images that could be for making 3D models. Doesn't look like a good idea to me. Make a drawing or just get started sketching out some model in a program.

>>24342 >Make a drawing or just get started sketching out some model in a program If I could draw I wouldn't really have much of a need for AI-generated art.

>>24360 I meant sketching. I don't understand what you're trying to do. >engineering blueprint pics, or something more like concept art or reference images that could be for making 3D models. You wont get around learning something. Sad but true, it's a lot of work.

>>24361 >it's a lot of work. Or a lot of waiting. I am a very patient man.

>>24191 Trying to make a picture without hair and neutral background, for using it in as head model.

>>24386 IMO this one seems a bit high on the encephalization quotient for a mature female, but is otherwisee rather appealing in general. Good progress Anon.

>>24386 So, I created some seed and prompt where one could start. > seed 4966549000887420 > skinhead female head with doll eyes like Alita but absolutely no hair, neutral and simple unicolor background in one color without any gradient >>24387 >The amount of brain mass exceeding that related to an animal's total body mass. Yes, this seems to be a problem. Maybe I can get that down a bit with future results. It took me some rounds to get the model not make any hair. But I guess it still interprets the size of the head with hair as the size of the head and then the hair would go on top of that. I'm using this on that site here: https://3d.csm.ai - which is for making 3D models of 2D pictures. It was in one of Matt Wolfes recents videos. But he said, he thinks it might not be available for free for a long time, since getting a result takes hours and that indicates a high amount of compute being necessary: https://youtu.be/rtXE9Knszws?t=17

>>24387 >encephalization quotient I tried. But it doesn't want to do it, at least not with my current seed and example pictures: > skinhead female head with a low encephalization quotient, small cranium compared to the eyes, a sized down forehead compared to the eyes, the forehead should end not far above the eyes, the face should be looking a bit like Summer Glau, with doll eyes like Alita but absolutely no hair, neutral and simple unicolor background in one color without any gradient negative prompt: lowres, high forehead, high encephalization quotient, big cranium

>>24386 >>24387 >>24392 I wanted to create a face from scratch, but when it comes to AI-generation I don't trust it to create something that can make a perfectly realistic human. And while I do like anime aesthetic, I personally just don't think it works well in 3d with realistic lighting. I'd like to make a face that's more of a composite image of attractive-looking faces, then tweak the color and texture to look like a good cosplay of an anime character, rather than base the head directly off of one. My problem has been finding decent composite heads that look around 18, since most composite faces seem to look a bit too old.

>>24395 >perfectly realistic human Do what you want, but I think big eyes are a extremely important factor. Also, this can help to make the point that she's not human, and we don't try to trick anyone. Making a "perfectly realistic human" was never my goal. That's a bit like the claim that chatbots or AI girlfriends should have a background story like a human and behave like a "real" but not real fantasy girlfriend. >personally just don't think it works well in 3d with realistic lighting. I think the AI generated pictures of the big eyed females I'm posting and the Alita movie shows otherwise, though this might be a matter of taste, of course.

>>24397 >Do what you want, but I think big eyes are a extremely important factor. Also, this can help to make the point that she's not human, and we don't try to trick anyone. I get you like the bigger-than-real eyes, but as for tricking people, why does that even matter? The whole point of the robowaifu is essentially to trick yourself. >Making a "perfectly realistic human" was never my goal. That's a bit like the claim that chatbots or AI girlfriends should have a background story like a human and behave like a "real" but not real fantasy girlfriend. Making a robot that can pass for human is an end goal of /robowaifu/ in general, because there's seemingly nothing that can't physically be done once that's accomplished. >I think the AI generated pictures of the big eyed females I'm posting and the Alita movie shows otherwise, though this might be a matter of taste, of course. AI generation gets a lot of subtle things wrong. I don't think you'd really get a true sense of how it'd look until you've got something physically in front of you. Personally, I've been working under the notion that something like a realistically-proportioned head is simple enough to get right, but the movement of the face is the most difficult part. As far as I've been able to tell, with the Alita movie they took the real actresses face and motion captured her expressions, using that to control the flexes of the CGI head, which was her own head given different proportions. Trying to make a head from scratch and get good-looking movements out of it seems trickier than having a real face to base it off of.

>>24402 >The whole point of the robowaifu is essentially to trick yourself. Nooo, having some good looking companion. >Making a robot that can pass for human is an end goal of /robowaifu/ in general One possible goal maybe, but not a very important one. This would be an extreme optimization for something unimportant. I'll leave that to other people. We had and still have guys here who are at least fine with robowaifus looking somewhat robotic or even want that, part of the target group are robosexuals (/clang/), others are open towards some /monster/ elements, and everyone likes catgirls. Then, I'm the one who strongly emphasized I wanted something more human-like than the /clang/ guys, but I certainly never wanted something realistically human as end goal. The robowaifus should look better than human females, and maybe kinda cuter without looking like human minors. They also might have quite some quirks and are more trustworthy in regards to accepting their role as property of someone. They don't need to pass as humans and maybe this would even be more offensive or disturbing to some people, I mean if they can't tell quite fast that something is a gynoid. >seemingly nothing that can't physically be done once that's accomplished. They will be different. Advantages and downsides, and we'll can go on working to reduce the downsides. Especially making them resilient, long-lasting, flexible in doing tasks, and energy efficient. Maybe self-repairing and many other things, all much more important than optimizing for making them looking and behaving like they were humans. >but the movement of the face is the most difficult part. I agree with that part to some extend. But Cameron for example mostly had reduced facial expressions and anime waifus also don't have a human-level amount of facial tissue movement. Also, some smiling and a few things are not that hard, I think: https://youtu.be/Esw5gjrFL-w >Trying to make a head from scratch and get good-looking movements out of it seems trickier than having a real face to base it off of. You're overthinking things. We'll try one step at a time. It's more an issue of (reliable) mechanics.

>>24403 >One of my first fox girls How did you find that? >Waifus are meant to be attractive companions Correct, it bothers me that so many don't get it like you do. I for one do not want my waifu to be too human. I just don't want a human woman, I want a robot. An anime girls face is desirable, though it will be terribly difficult. It is however, a worthy endeavor. Picrel is something I want to turn from and AI generationg into a real waifu.

>>24416 The picrel is a repost. I have a folder with memes and pictures, I often don't remember from where I got them. In that case I did remember, but it fitted well to my posting. >not want my waifu to be too human Thanks. I'm even the one here who was pushing the most towards human-like looks (~Cameron with Alita-eyes). But I'm often running into people on other platforms who think that to get companion bots as alternative for women we would have to build something like the bots in Westworld. So that would be the goal and maybe anything before is just toys and prototypes. Then often based on that, the conclusion is that this would be going to take decades or a century to get us there. I think we went /meta/ here, btw. >AI generationg into a real waifu. Yeah, looks nice. https://3d.csm.ai doesn't work well with heads so far. The previews don't look very good, and it takes a few days to get a result. I think they're manually checking the results and only hand them out if they meet certain standards, so people won't just show off something bad looking and telling where it is from. Either that's the reason for the delay or too many people found the service and there's a long line.

>>24416 This channel here made a bunch of pictures based on the female characters in Chainsaw Man. Haven't seen it yet, but the comment section seems to think that they look like the original would look in 3D: https://youtu.be/XEYIkDleJo8

>>24340 >Nooo, having some good looking companion. The companion part is the part that's a trick. It's an appliance, not a companion. >not a very important one. To you. >maybe this would even be more offensive or disturbing to some people That also applies to the concept of robowaifus as a whole. Those people do not matter. >anime waifus also don't have a human-level amount of facial tissue movement Depending on the anime, they can have significantly more. >Also, some smiling and a few things are not that hard, I think I'm looking for at least Source Engine levels of facial expressions, where some subtlety is possible. Not some Thomas The Tank Engine, brick-to-the-face levels of exaggeration. >You're overthinking things. We'll try one step at a time. It's more an issue of (reliable) mechanics. Not really. I plan on hiring someone to model and rig my robowaifu (and a skull) for me, using references I've made from a composite of different faces. Puppeting the eyes, jaw, and the skin will be compared to flex animation of the model, so making real facial expressions would be no different than doing it in real time in a videogame. >>24416 >An anime girls face is desirable, though it will be terribly difficult. It is however, a worthy endeavor. Anime translates to 3D very, very poorly. At best it might look like a life-sized anime figurine, at worst it's like one of those creepy "Animegao kigurumi" cosplays. And the former would probably look pretty lousy when animated.

>>24416 >Picrel is something I want to turn from and AI generationg into a real waifu. She would be a charming robowaifu Kiwi. I'm glad that the consensus here generally seems to be coming down on the side of 'animu aesthetic, not realism', b/c IMHO we are a loong way off from being able to match the subtleties involved merely in general body motions, much less the far more difficult task of facial realism complexities. We'll get there eventually, but it will take decades before we're likely to achieve Westworld-tier robowaifus implying any man would want to once we all see how great animu robowaifus are! :DD. >>24494 >It's an appliance, not a companion. Haha, so a My Dear Marie proponent heh? :^) Having a waifu is practically a spiritual circumstance for many anons, Anon. It goes far deeper than just the physical manifestation. For us as robowaifu engineers, this is a plus. Makes our jobs easier actually b/c the mind fills in many of the 'blanks'. >Not some Thomas The Tank Engine, brick-to-the-face levels of exaggeration. NGL lel'd. :^)

>>24340 >Haha, so a My Dear Marie proponent heh? :^) I know nothing about it other than what Wikipedia says the story is. >It goes far deeper than just the physical manifestation. I'm a hylic, so whatever floats your boat.

>>24506 >we are a loong way off from being able to match the subtleties involved merely in general body motions, much less the far more difficult task of facial realism complexities But we will get close enough for even the people who care about a high level of human-likeness. I'm not claiming facial animations are going to be extremely hard, but I agree on the subtleties. More tiny muscles will be an extra effort. Fine tuning for making it absolutely realistic would be a difficult optimization on top of that. But also, think of getting rid of every mild noise. Literally perfect emulation of humans would be a very special fine tuning task, while it's not really necessary. >it will take decades before we're likely to achieve Westworld-tier robowaifus Maybe, my point is rather that it's not worth the extra work for doing that kind of optimization. So maybe no one is going to work much on it, but I don't know. We're trying to harvest the low hanging fruits here, and even that is very hard and not many people are working on it. Energy consumption, energy storage and noiseless motion would be the main problem if we wanted to go Westworld realism. In reality, our waifus will mostly be stay at home wife's and will have a plug for charging. I will likely do other things when we are there, not working on making them more "realistic". If I still have time to have children (sons) through surrogacy, then I will at some quite early point work on child care skills in regards to older and older children, not making the waifus looking like real humans. We're went /meta/ a while ago, btw.

>>24516 >I know nothing about it other than what Wikipedia says the story is. It was just a lame joke on my part. But the point about 'a robowaifu being nothing more than the equivalent of a fancy washing machine' actually works in our favor here, within the marketplace of ideas. To wit: "Why should feminists women get their pantsu in such a knot over just another washing machine?" :^) >I'm a hylic, so whatever floats your boat. The point isn't me or my preferences friend. :^) In fact it's not even you or yours... it's the reality on the ground of millions of men without loving companions today b/c of the literal Satanic evils of the Globohomo Big-Technology Government we're all burdened in dealing with. That's why we're even here on /robowaifu/, as far as I'm concerned. >>24519 Yes, you're totally correct NoidoDev. We will get there with rather high-fidelity kinematics & motion control, as an overall effort in the end. Likely, I focus a little too much on the difficulties and should stay more upbeat instead? :^) Regardless, I certainly agree with you that we should not be attempting to create perfectly lifelife robowaifus. I was merely addressing the consensus vs. the realworld engineering difficulties. >We're went /meta/ a while ago, btw. Heh, good point. Probably need to do yet another tedious relocation. :P BTW, that image is an excellent one IMO. Nice choice, Anon. Cheers. :^) >=== -prose edit

>>24340 >To wit: "Why should feminists women get their pantsu in such a knot over just another washing machine?" :^)

>>24397 Weird question NoidoDev, but do you have a specific size preference for bigger-than-real eyes?

Yumemi Hoshino via SDLX on stablecog.com Seed 343045970 I used another picture as pattern. At first, one from the anime. I had to use negative prompt "no text" to get rid of the text I had for the meme. Prompt: Photo-Realistic picture of a doll-like female with realistic Alita doll-eyes - red necktie - green eyes - light blue hair - the loops to the side of her head are a bendable display or fabric no metal or tubes

The latter ones are inspired by one of Sukabu's pictures, a gynoid repairing herself, I tried to make the model reinterpreting it into something in 3D.

>>26225 >>26226 Really nicely done, Anon. Gorgeous eyes in particular.

This channel here does quite a good job on AI generated females. Though, they're mostly quite east Asian looking, which is fine but maybe a bit monotonous. I'm also not watching this a lot, but I appreciate the quality and look into it from time to time. https://www.youtube.com/@eou_of https://www.youtu.be/oUunsiA1D6U https://www.youtu.be/t1V6Rvhx-xY

Tried to help the Emily dev to get some ideas (on Discord)

>>26277 the eyes are too creepy it looks like a pair of evil eye talismans glued on

>>26278 I like them or some of them, at least as a concept. But I generally realized with all the AI generated images I'm making, that I'm quite flexible when it comes to what I perceive as good enough looking.

>>26281 no their good its just the eyes, i think eyelids are obligatory to avoid that uncanny valley feeling

>>26282 > i think eyelids are obligatory to avoid that uncanny valley feeling Good point.

Robowaifu Fairy Princesses Need Love & Rescuing! PLS SEND HALP ANON! :^)

>>26450 >Halp Mai waifu is here to halp us! Mixing in Nun LoRA's can help get better crosses.

>>26501 A beaut!! :^)

new to the thread. not running local SD. currently using https://perchance.org/ai-text-to-image-generator for generation. vintage anime setting or neon vintage anime returns really good results due to (imo) the 80's and 90's having a LOT of really good mecha/robot/android influences from things like iczer-one, bubblegim crisis, bubblegum crash, plastic little, armitage III, battle angel alita, etc. Newtype magazine from the 90's showcases a lot of great reference art. been posting various word prompt renders on the emmy the robot thread. Having a shit of a time getting robot skin/metallic skin/whatever. Recommendations that might help me get closer to the Emmy/nandroid theme would be very welcome since, as it was pointed out, most of my generations are quadruple amputees with robot limbs but a human core body :)

>>27018 Thanks for posting in the proper thread. I recommend adding "skintight", "bodycon", and "joints" they can be helpful towards matching the Nandroid aesthetic.

from a prompt standpoint, how 'gathered' do concepts and sentence structure need to be? Are there notable differences if you spend 3 sentences on environmental descriptors and then spend 3 sentences describing the character and another 3 or 4 sentences explaining emotion in a setting or a character? Does it matter if you jumble the order. environment, emotion, character? Guess what i'm after is....how do you trigger priority and 'sticky-ness' of words so that you get less wildcard returns?

>>27035 Nevr, ever write a sentence in a prompt field for Stable Diffusion. Prompts are composed of keywords with modifiers. You can use the same word multiple times but, using an elipse around the word is both cleaner and more effective. Example: 1girl, maid dress, robotic joints, blue eyes, black hair, choker, wall, Would work better to get your picture compared to; A maid with blue eyes and black hair in the living room with robotic joints. This second example would now include "A", "and", "with", "in", "the", "with" as prompts that are going to distort your results. You're painting with words, not sentences. You also want to use as few words to gain your ideal result as possible. This allows each keyword to have greater strength, giving you greater control. You also gain an easier base prompt to adapt into other images. I save my favorite prompts as text files to reuse and reference later.

>>27036 I was thinking along the similar lines of 'paint with words' as well. The prompt spits out some great passages (lulu being found in a pile of garbage) but....yeah, I'm noticing that brevity is best when it comes to getting a prompt to stick. Another thought - are......obscure words somehow not triggered from an descriptive noun or adjective standpoint? I might use 'lithe' or 'svelt' to describe something but its very hit and miss on art generation

>>27035 >>27036 As you can see, the first one is what I wanted. The second one is lewd and I don't understand why. (I tried it a few times, they were all lewd, I really don't know why.) By adding excess keywords, I lost control and precision. (Model used was 7th_anime_v3, no LoRA's, generic negative prompt.)

>>27038 These generations are based on the associated keywords of the model and LoRA used. Because of this, it is best to use common words, as they are more likely to have been associated with training images. Instead of "lithe", I recommend using thin. As you can see, "lithe" caused my maid to become deformed, "thin" caused her to be lithe. I highly recommend experimenting using one term and a well know subject, such as "maid" to see how that term modifies your subject in isolation.

>>27036 Very helpful information, Kiwi. Any chance you could create a post-sized primer for 'babby's first AI arts'? :^)

>>27048 is there a running list of nan-nouns,nan-adjective, and nan-key words that i can use for more refinement? while i love the 80's/90's anime vibe of the renders, Emmy is more closely inspired by american comics/oldschool disney/vintage pop media of the time

AI generated images of RobowAIfus

There's already a thread for this ( >>17763). These are some nice images, though.

>>29334 Much as I really value your selections, OP, Greentext anon is right (>>29349). I"m going to have to merge your thread into the other soon. Two reasons: 1. We really value keeping threads on-topic here. The catalog is your friend. 2. We want to keep our limited thread budget under control, and we're a board that retains focused, long-running topical threads. Hope you understand, Anon. Again, nice choices!

>>29353 I completely understand, I actually looked in the search option for AI image but I didn’t get a result back so I thought there wasn’t a thread like this already.

>>29359 Yeah, my apologies. I may edit the subject of this thread to better encompass such a search. Thanks for your understanding Anon. Cheers.

>>29334 The 'baby on board' and the 'catgrill inna box' are pretty iconic images, IMO. >>29335 Also, the 'Holly Homemaker' as well.Good marketing material!

>>29378 3rd pic would be a great image for a /wbg/ thread. (cf. >>29313, et al) >>29379 Also, the final two here as well.

>>29342 >>29343 Okay, this is getting really good. (I'm ignoring the fat one). The first in here >>29344 is also great.

>>29343 curvy robowaifu best for laifu more curvy robowaifus!

>>29344 That face speak volumes.

Happy Valentine's Day! We're another day closer to your waifus being real! Until then, if you do a little drawing, it can really make a difference in how your art ends up. I think it's a beautiful thing to see something come to be, through the mutual efforts of an Anon and his machine.

Title of thread is wrong, should be something like AI Robowaifu Art. Stable Diffusion is not the only AI image generator, all of the images I posted in this thread were typed in DALL E 3

One of my streams of pics is accessible online, btw: https://stablecog.com/NoidoDev - This includes the ones I didn't download and wouldn't spread further, though.

>>29610 Nice but maybe a bit too creepy looking, since the machine looks too much like a machine. I contemplated adding babies to my waifu pics.

Looking for ideas on how I can make the (picrel) antenna less terrible-looking. It’ll be movable with cables and anywhere from 2” - 5” in length, and I was planning to add LEDs so I can make it a part of her “face”/tied to her “emotions” My current idea is like the attached SD gen, with a centered antenna in the middle of her head. Of course, I personally have zero attachment to humanoid aesthetics and whatnot, so I could use advice from people who *aren’t* completely faceblind/function-over-form. I can move the outer antenna/conductive rings into the body/stem of the antenna itself, but the LIDAR block on top (2.3”x2.3”x1.7”) is pretty much required. The upper part (the cylindrical “dome”) cannot easily be covered, either. (apologies if this is the wrong thread for aesthetic brainstorming, couldn’t find one more suitable in the catalog)

>>29644 The first thing that comes to mind is a top-hat. Maybe as the top of a steam-punk style with the goggles concealing other sensors. Or as part of a mini top-hat, anime style. depends on what style you are going for.

>>29644 Hmm. If you're trying to stick as close to human as possible, the only thing I can think of that would look nice is some kind of non-removeable headwear. A crown or tiara, perhaps? The LIDAR device could be mounted in the body, and the dome protrudes from the top.

>>29644 >>29645 Taking the top-hat concept a little further, if you want the techno-body like in your pic, you could go for a transparent hat so the antenna is showcased as part of it, hiding in plain sight. You could use a transparent plastic container cut off and inverted. Mold a hat brim form with modeling clay and vacu-form a transparent brim from plastic sheet over it and glue the two together. Use magnets to hold it in place on her head so she could raise/tip her hat when appropriate to add character. Include a little slack wire for the purpose, with a connector for complete removal if desired.

>>29645 Can any of the lower part of the antenna be inside her head? If so, and her hair (if any) isn't enough to conceal the rest, you could go for the "headphones" look like in your picrel, with the protruding emitter/receiver as part of the headband design, possibly with other antennas on the sides or folded up boom-mikes. See picrels. Other than those all I can think of would be as part of a restraining band for a hairstyle like a puffed ponytail, see picrel

>>29649 Wrong crosslink, should be >>29644 Mod, please fix and delete this. Thanx.

>>29645 >>29647 >>29648 >>29649 Damn, I hadn’t even considered making it a tophat or another accessory— much appreciated, thanks guys. I’d only been thinking about mecha or monster-girl aesthetics… and as for my design goals, I’m not at all attached to ”near human”— in fact, I’d prefer she be quite obviously robotic, 100% function-over-form Anyway, functionality-wise, the antenna needs to be able to “sway” around enough to let the LIDAR “peek” over the head so it can scan the ground immediately in front of the robot (worst-case, of course, but that always needs to be accounted for in design). The antenna head (LIDAR dome) itself needs to be the highest point of the robot (so the scan plane isn’t obscured); presumably, that implies it’s centered, but it doesn’t *need* to be. During normal operation, I imagine it’ll sway by 0.25-0.5in as it angles the scan plane to hit areas that the optical sensors marked “uncertain” (stereoscopic correlation -> depth failed). Anyway, I wanted to make it something that looks “natural”-ish moving around a bit during operation, which is why I defaulted to an antenna. I might be able to do a “dancing tophat” or similar theme too, tho… Frankly, I don’t fathom what normies will or won’t find cool/scary/repulsive, so I figured I should ask before going ahead with that part of the design. Worth noting that I’ll likely do “inhuman” ears as well, perhaps headphone or fantasy style (elf/dragon/fox/etc.?) I personally don’t mind having a “chimera robo-girl”, but perhaps people would find a cohesive “theme” more pleasant? Plus, I plan on having a fairly advanced audio localization system, so the orientation (and ideally the shape) of the ears will be dynamic

>>29695 Hmmm, asymmetry will be tough to pull off aesthetically. I think the picrel of the girl-bot is an example of it not working. It works for Robby because even though the antennas on each side are at different angles, they are the same shape and there are two of them, giving balance. For having the LIDAR unit tilt, you could just have her head tilt to "glance" downward, just as we meat-bags do.

>>29695 I’m not at all attached to ”near human”— in fact, I’d prefer she be quite obviously robotic, 100% function-over-form Wow, I completely misinterpreted your original post, to the point of essentially reading it backwards. I thought you wanted her to stick as close to human as possible. If we throw that out the window, the possibilities increase significantly. First, I'll get what seems to be your primary concern out of the way: >Frankly, I don’t fathom what normies will or won’t find cool/scary/repulsive, so I figured I should ask before going ahead with that part of the design. What normalniggers find to be "okay" and "attractive" is little more than vogue - it changes over time, and we are in a period where vogue changes quite rapidly. If you try to stay in-vogue, you'll just get left behind. Don't waste your time chasing someone else's vision of beauty. Build a waifu who is beautiful to you, and those who agree with your designs will come to you. Now then, I have many more ideas for how you could mount your waifu's LIDAR system in a more attractive manner, but I'll stick to basic overviews based on a handful of general aesthetics: High fantasy: The LIDAR block could be mounted onto the end of a horn, antler, sigma (botanical), or halo. There are other possibilities of course, but I can feel this post becoming quite long as-is. Concerning the horn: it could either be a singular horn (unicorn-like, protruding from the forehead), or the largest horn of an asymmetrical design (you'd be surprised (or maybe not) at how good asymmetrical horns can look). The horn itself can be hollow, and made from lightweight plastic. This gives you room for tilting in the LIDAR device. For instance, the antenna could be within the body of a chonky horn, and only the dome protrudes from the top. Tilt could be achieved by mounting the antenna to tension cables, which shift the block on the top when articulated. An antler based system would work the same, I surmise, just with a different form-factor. Concerning the stigma: if you don't feel like looking it up, the stigma is the endpoint of the pistil (stalk-like thing in the middle of the flower). This would necessitate that your waifu is botanically themed, but it can be done. It'll also look more natural than the horn idea, since it's easier to pass-off the movement of the LIDAR device as normal "swaying" of the "plant". Concerning the halo: there are many forms a halo can take, but they all basically amount to you adding plastic mass in a vague ring or circle shape to your waifu. The idea here is similar to horns or antlers, just implemented differently. The LIDAR block would likely have to be static in the halo, as the entire halo structure articulates. Elves, animal-ear autism, and (you): Every feasible ear design would work roughly the same way, in that the LIDAR device would be integrated into an articulating "cartilidge". I'd personally recommend felinid, canid, or lagomorphic (upright, not floppy) ears, since they're large and allow many degrees of freedom while allowing the LIDAR device to be as high-up as possible on the structure. I don't know too much about LIDAR systems, but I assume that they work best when located as high-up as possible, so I'd personally stray away from ears that stick out the sides, like in elves. I could easily be wrong there however, since I'm not concerning myself with LIDAR (I'm personally leaning more towards SONAR). Sci-fi: There are two primary thoughts I have that could work with this aesthetic: The first is that you lean into the futirist aesthetic, and let the LIDAR device shine on its own. The second is that you merge this with fantasy, and create sybernetic "horns" or something similar that look more metallic and technologically advanced. The following ideas are ones that can work very well, but only if you really commit to the aesthetic. I don't reccomend them unless you have an eye for them. Dark fantasy: Similar ideas to fantasy, except focused more on monster-like and undead aesthetics. Wouldn't a lich waifu look nice with a pretty bone crown or strange crown-like artefact fused to her head (the crown jewel of said artefact being the LIDAR block)? Concepts like this can look pretty, and I've seen it done very well, but it's an acquired taste for most. Dark sci-fi: Similar to sci-fi, just messier and with old school (typically) sci-fi dystopian elements. Perhaps give her some complex-looking visor headgear made from dark metal and have some (fake) wires stick out? This, too can be fused with fantasy elements. Eldritch: Are you looking into the abyss, or out of it? Scrape away at the blackened dregs which have settled upon the bottommost depths of your imagination, and form them into a vision of terrifying beauty never before seen by mankind. These are just a few basic off-the-cuff ideas. Regardless of what you choose, you'll want ot theme your entire waifu accordingly. I'd advise considering her appearance as a whole, instead of focusing only on parts. That head has to be attached to a body, after all.

>>29695 >I’d prefer she be quite obviously robotic, 100% function-over-form OK, that narrows it down a bit. How about an "up-do" hairstyle made of wiring, heat pipes, fiber-optics, jewels(camera lenses), etc, in a style similar to that of yeoman Janice Rand in Star Trek TOS, with the LIDAR box/dome on top? See picrels. This could conceal the tilting mechanism, and the rings could be hidden or exposed as part of the do. The sloping shape would allow a clear field of view all around. You could have the mood LEDs light the fiber optics. Fiber optic "hair" could fall behind to shoulder-length or longer with more metal ornament (tinsel?) suggested by the hairstyle of Jennifer Connely in "Labyrinth". Black and white Rand pic with more silvery look. A feminine but very artificial look, as functional as you can make it. Sorry, I don't have any robot art with anything close yet.

Generative image making isnt very good for designing ideas for 3D objects since it can only create identifiable objects already commonly drawn and photographed before. Otherwise details get melted into unidentifiable masses if you prompt anything but a cliche or lots of small parts are used and not enough people draw it exactly the same. In many cases you could have found images of already drawn by someone relatively easily with some ideas more clearly, just not always in a single image. All you get is a seemingly polished image but is actually full of flaws in any detail so it only gives a vague idea but not detailed like even a simple sketch can better convey sometimes. Id say it is a waste of processing power and time. >>29699 >>29698 >>29695 If you mean a LiDAR that is 360 you can place it in the neck behind translucent plastic or resin that doesnt block IR. Just need short enough hair and a way to move the head that wont block LiDAR too much which might be challenging though. What if it was a staff she carries or even just a simple tower or rod protruding from her shoulder or back above her head? or what about a small drone that automatically tracks her that LiDAR and sends the info to her guiding her third person perspective but is technically part of her? When it comes to stationary LiDAR that is way easier but you would need a minimum of two to four depending where placed. I had posted some more affordable LiDAR in a few posts in the vision thread if anyone is interested.

I've been generating robot girls (among many other things) using A1111. I'll dump my stuff in here. >what checkpoint? My checkpoint is the pony one and I swap loras around all the time but I usually use anime-themed ones. >can I get a .png version? If you want a higher definition of one of these pics, let me know and I can drop the .png file. >what prompt did you use? Most of these utilized PSO as a tag with various other things for posing and backgrounds. If you want specific prompts I used for each set, let me know. I don't know if this board wipes metadata, if it doesn't you should be able to pull it from the pictures. >can you make "x" themed art? I can try, but I'm no professional prompter. If you want me to use another checkpoint or a specific lora let me know and I'll look into it. >Can I use this art in my project? Nothing I can really do to stop you. It'd be polite to at least note that you didn't make it and you pulled it from the net but I don't need to sign my name on any of this stuff. >Some of these gens are kinda shit... HIT THE LEVER! You never know what you're going to get. If all 4 outputs shit the bed, I won't be inclined to share them but if 2 out the 4 are decent I'll post the grid.

>>30938 These more generic bots were made without the PSO tag. I added "nuse" and "hospital" for some of the nurse bot pics. Big tiddies too.

>>30939 These were some of my earlier outputs.

>>30940 I'd deleted the grid versions of these because I was afraid they were taking up too much space. Also, all these spoiled ones get pretty lewd. >Why are the robots creating milk? Why not?

>>30941 I forget what the prompt was for these bigger bodied waifu bots.

>>30942 I for some of these lewd ones, I was experimenting with how the model would create "robot pussy". I'm not sure if that's the actual tag I used, I can dig into it as necessary.

>>30943 Some of these got really weird but I liked the designs so I held onto them.

>>30944 sorry for all the robot puss. I wanted to see what the SD model could do.

>>30945 Sorry for all the slop. Hope you enjoyed my generations. The last one I'm not sure about being NSFW but someone might construe her color scheme as showing nips so I hid it.

>>30945 They're great. Interesting what it came up with.

>>31097 What's cool is you can blend existing characters with robot features for some unique outputs. I've been wanting to practice inpating and some other tricks to touch up generations but I haven't had time to as of late.

>>31093 I'd like to please order 5 dozen, thank you. :D

Some lulu for the thread. Using jennifer connelly from labrynth as reference

via hedra.com

>>31694 >>31699 Lol. You know NoidoDev, in a few years from now when it's not just virtual waifus, but the real-world Model A robowaifus and others begin appearing, and tens of thousands of Anons (at first, then later millions of Joe Sixpacks) all suddenly realize together (roughly simultaneously) that their own, real, robowaifu is within both their time and financial means -- usurped legislators be damned -- that we're all going to see some serious things begin happening globally. >tl;dr It will be a case of the proverbial "20-year, 'overnight' success". Cheers Anon. Thanks for sharing these. :^) <also, Alita best waifu ofc. :D >=== -prose edit

>>31700 > 20-year Oh my, this would be too long. I hope I can snap out of my current phase soon, and get to work more on it. Anyways, for better virtual girlfriends it would still be necessary to extract "graphemes" (https://phonicsandstuff.com/articles/what-are-graphemes ) and make this way faster by translating it directly into an animation. I guess this would be much easier with animated waifus in the first place. Code needs to be able to generate facial expressions very fast. These models her will be good for content creation, but for conversations it's still to slow, and needs too much resources. >>31701 "Alita" here was rather interpreted as fat than old, and I need to be more careful about what I type to avoid errors. >=== -hotlink patch

>>31702 >Oh my, this [20yrs] would be too long. Don't worry. The clock started back in 2014, when we began these threads over on 4cuck/g/pol/b/, remember? We should see working, initial prototypes within 4-5yrs now, I'd recken. :^) >I hope I can snap out of my current phase soon, and get to work more on it. I hope you do too, Anon. I'm praying for every productive anon here to finally succeed with our collective dreams for robowaifus. Keep moving forward.

>>24392 >>24394 Just tested FLUX.1 on this, and it doesn't want to reduce the forehead or cut it off. But maybe some are okay, and in the last one it was just the wrong prompt. I don't have the prompts anymore, forgot to save them. Otherwise this has to be done manually then fine tuning at some point. >Why again? No hair means a 2D to 3D model might work.

>>32638 I tested it with meshy.ai and it looks okayish from some angles, but it's ultimately botched. Also, not close enough to the original. I think it is related to the bad take on the big head. The eyes would need to be higher, so the face doesn't need to be "long". It's a bad combination of anime and real, since in anime the make the noses more cat like.

>>32653 I'm trying more with making the generated picture into a stone skull. This works partially better in the 2D to 3D stage, at least I got a better profile in one case. Though, in another one it went silly and made it into some ugly male. In the first image I made the mistake to let it have realistic eyes, which looks nice but this is exactly one of the problems for the 2D to 3D transform, since it has eye lashes.

>>32638 >>32653 >>32654 Neat! A few of these look really good, NoidoDev. Please keep up the good work. Cheers. :^)

>>32663 Thanks, I might upload them at some point to my Mega account, especially if people are interested. I think, often it will rather be useful to use half of the head and mirror it, because the other side is botched. These 2D to 3D models really don't work very well. I also tried text to 3D (last picrel), but this works even worse, though I might not have tried hard enough.

>>32668 Here some prompts to make busts in Flux.1 > stone bust with a very smooth and clean surface of a hairless female head with a low encephalization quotient, small cranium compared to the eyes, a sized down forehead compared to the eyes, the forehead should end not far above the eyes, bigger than normal eyes, smooth skin and lips, no eye lashes, no hair, no hair in the face, no hair around the eyes, no background, no gradient background colour > isometric view on a hairless stone bust with a very smooth and clean surface of a hairless female head, with a low encephalization quotient, wide face, high cheek bones far away from each other, very big eyes, smooth skin and lips, no eye lashes, no hair around the eyes, no background, no headgear or cap > a hairless stone bust with a very smooth and clean surface of a hairless young female head, with a low encephalization quotient, wide face, high cheek bones far away from each other, big eyes and a wider area around the eyes, smooth skin and lips

Some other 3D models worked also well, but I had to struggle a bit at first, since it made very bland and cute play toy dolls instead of silicone ones. So if you mention something like a doll in the prompt, you really have to tell it you want every part to look realistic. Also, in this case it needs to be a "youthful adult" or it will make it very cute. This here should work: >realistic portrait of a youthful adult, with very big but realistic looking eyes with a realistic eyeball, very realistic redblonde hair, realistic eye brows out of human hair, light eye colour and realistic looking skin, very smooth lips without folds I used this, btw: https://huggingface.co/spaces/multimodalart/FLUX.1-merged - but FLUX can also be hosted (with 16GB 8bit quantized or 24GB).

>>32671 Interesting research, hope it'll help us create more beautiful waifus. 1st one packin' heaps of cute.

>>32673 Some of these look remarkable, NoidoDev. Can you get the thing to spit out the bust's geometry mesh? If so, then I could maybe make an effort to retopo one or two of them.

>>32679 >Can you get the thing to spit out the bust's geometry mesh? If so, then I could maybe make an effort to retopo one or two of them. Yeah, it would be easier if Meshy would allow publishing on their platform, they have the option but this is currently not enabled, maybe because they would need to moderate it. I downloaded some of them, and might download more and I can upload them to a Mega folder. The board here doesn't accept most file types. Meshy allows for many different file types to export and download, which one would be the most recommended? blend, obj, fsomething, glb, stl, and maybe one more?

>>32680 >which one would be the most recommended? blend, obj For myself, I'd do any editing in either Maya or Blender, so yeah either of those would be good. And .obj is probably about as close to a universal DCC file format as we'll have any time soon.

>>32680 >>32681 If .obj files can be made, I can clean it up and make some changes if wanted. I've had troubles with facial modeling, so working together on this would be great.

>>32673 These are really fantastic. In this set I linked the first to me is cute but I like the last picture best. I think I like the large but not too large eyes, and they are more horizontal than the first picture.

>>32683 Sure, Kiwi! I would absolutely enjoy it if we built a team to collaborate together on these type things. From my own perspective on this matter, I'd primarily be interested in Quad-based retopology for any generated mesh by such tools as the one NoidoDev is using. Both Maya and Blender have straightforward tools for such operations. This will allow for smooth deformations of such meshes during subsequent skinned animations. Beyond that, I think taking a hard-surface modeller's approach to the robowaifu systems will serve us all well for the large majority of robowaifu components, such as: * endoskeletons * connectors * servos/actuators of most types * gearing * internal components (cooling, batteries, etc.) * joints * external shells * many more In fact, about the only things that would need organic modelling approaches are the soft 'skin' (silicones & the like), underlying soft 'tissue' (LARP foam & the like), and flexible, moving bits like Bowden cables, wiring harnesses, etc., on the insides. Many of these latter are probably also primarily best modelled hard-surface as well. I apologize if I'm getting ahead of myself here, just wanted to throw those concepts out there in advance for discussion (which BTW, we should probably continue in a different thread than this one haha). :D Cheers.

>>32676 > I can clean it up and make some changes if wanted. I've had troubles with facial modeling, ... You can look here what can be done quite easily: >>24484 (carving out the eye balls, and I could remove the ears - Btw, I posted the 3D AI rendering topic here in the "art" thread, but should've put it over there in the 3D rendering thread anyways. Also, I wanted to use CSM again like I did before, but ended up trying Meshy because I couldn't log into CSM. This website might still be worth a try though. > ... so working together on this would be great. I'm often just pitching ideas and pointing out a path. We simply don't have enough people. My take on the face still is: - Option One: Take a available model from some 3D website. - Option Two: Make one, open source it or don't. - Option Three: Personally I will most likely pay somebody to make the face for my first robot girl if I can't do it with AI by then. If I make this freely available or not will be determined by me seeing this one as my Waifu or just some standard face. Idk yet. - Long Term: The tech demo here is meant to show what can be done with AI right now. I guess it will get better. I have some ideas how to make my own model specifically for 3D faces. We should have something going from 2D to 3D, so every bot can have a different face. - Short Term: I might be able to make a simple head in OpenSCAD where people could attach a face onto. It would need some adjusting of variables to make them fit. Either way, we probably should go on with this in the prototyping thread. (picrels are OT but shows the state of what can be done, videorel are some waifu face videos, done with Stable Video Diffusion on HF)

I was looking down this thread and something occurred to me. The cartoon pictures with really big eyes look good but the realistic eyes of the same diameter do not. I think, it's because the cartoons are flat and the realistic are not, making the realistic look bug eyed. Just a thought. I wonder if the big realistic were more flat shaped parallel to the face if they would look better? How the hell you would word that to get a LLM to display it would be a tricky problem. This, as I said, is just about right. Really nice. http://bhlnasxdkbaoxf4gtpbhavref7l2j3bwooes77hqcacxztkindztzrad.onion/.media/ba9d90eaca0c8af53c3fd8ecdbc1c39aae1f8afa167551907f20934bf5a56466.jpg

This is not stable diffusion but a new text to AI modeler, FLUX.1, People say it's WAY faster and does great pictures. Some are raving about it. Hype??? The demos looked good to me. It's not so good for NSFW I gather. FLUX.1-schnell this version is able to be run locally, link, https://github.com/black-forest-labs/flux https://huggingface.co/black-forest-labs/FLUX.1-schnell you can do test pictures at links provided in the link. The one at, https://replicate.com/black-forest-labs/flux-schnell It was picky about NSFW. I was trying to get it to display female skeletons. Which is hardly unsafe. I got it to do a few. I wanted to get it to show muscles and tendons on one side and skeleton on the other. Not much luck. Sometimes it would display a skeleton sometimes not and finally it would only give me,"can't display NSFW. Sigh. I found this through another site that has downloadable SD modded to also do NSFW pictures, here, https://civitai.com/models/128607?modelVersionId=247444 I think...you have to have the BakedVAE version to run local or to keep you from having to install other stuff. I have no idea how to get these going but they seem interesting.

>>32728 >Stable Diffusion is in the headline of the thread, but this should be changed, since this thread always has been used for other models. Sure, that'd be OK, NoidoDev. Here's the current OP subject: >Stable Diffusion for Robowaifu Art What should it say instead in your opinion?

>>32695 >cartoon pictures with really big eyes look good but the realistic eyes of the same diameter do not Not sure what you are referring to, but I like the dolls with big eyes. The smaller ones are also fine, but more so from near distance. It's interesting that quite some people don't like Alita-like eyes, for me it's just fine or even great, though in her case it's really the limit. I also prefer these waifus to not look too much like humans, just somewhat. >>32698 >This is not stable diffusion but a new text to AI modeler, FLUX.1 Yeah, I mentioned this several times and linked it here >>32673 - Stable Diffusion is in the headline of the thread, but this should be changed, since this thread always has been used for other models. >It was picky about NSFW. People in some comments on YT pointed out that no base model ever was good for NSFW nor will there ever be one. People have to wait for the fine tunes. > I was trying to get it to display female skeletons. Interesting idea.

>>32730 Sorry, I deleted my posting while you were writing an referring to it. I moved the part about 3D modeling over to the correct thread, since any further discussion on 3D modeling should go on over there. >What should it say instead in your opinion? Idk, Robowaifu Image and Video Generation? It's even limited to images and maybe video, since we have another thread for "Propaganda" and anyone posting music would rather go there, at least I did. But should this even be two threads, or getting merged (which seems to be difficult)?! Idk. It will overlap with music videos, and then what about 3D files for games or something? We have a 3D modeling thread, but this is meant for knowledge on 3D modeling for building waifus, then there's also a prototype thread where actual projects and showcases should go. No one follows this perfectly.

>>32733 Probably best to limit it to image generation, but also tolerating clips. Anything more advanced goes into the propaganda thread. The 3D part of the posting and follow-up to the hairless sculpt to 3D is here: >>32735 - I also added some non-functional generated full body 3D model.

>>32736 OK, done. Let me know if that's alright or you want to change something. BTW, regarding: >"Anything more advanced goes into the propaganda thread" we're still waiting on a new thread, since the 'current' one still doesn't have the OP pic(s). Would you kindly just create a 'blank' thread with pics you'd like, and I'll migrate the related posts into it? Thanks, Anon! Cheers. :^)

>>32731 >cartoon pictures with really big >eyes look good but the >realistic eyes of the same >diameter do not I didn't explain it well. The key, I think, is the flatness of the cartoon. The eyes look flatter and do not bulge out like a real round eye would. So realistically proportioned LARGE eyes, bulge. Like Gollum. No Gollum waifus for me. >FLUX.1 >Yeah, I mentioned this several >times and linked it Apologies. I missed that or more likely just forgot. > I was trying to get it to display >female skeletons. What I really wanted was to have a skeleton on one side and the muscles on the other side. So you could see the relation of the muscles, and tendons, to the bone structure. I would expect this could be done if you had the right AI and lucked out, prompting it so it would regurgitate the right thing. It would be helpful I believe to have a series of pictures like this from many angles. If anyone can pull this off or has a resource, idea, etc. on how to do this it would be great. Another good set would be the same model with muscles, tendons, on one side and the full skin with the fat that makes up the model on the other, and from all angles. And the holy grail would be all of these with the model pre-chosen by the user and fed to the AI and it giving all these model break downs. I expect that would be asking too much.

>>32731 BTW the models shown in the link look great. I wonder if they would look Gollumy if from the side view.

Tried some more. Also tried to make some "realistic" looking selfies, which I saw in a video and made some people freak out. But the prompt wasn't shared, and mine don't have the right grainy look and the mirror isn't really visible. I thought it would be fun to make some realistic looking ones, but then with women that aren't looking realistic. https://huggingface.co/spaces/multimodalart/flux-lora-the-explorer >>32780 I could try in Meshy, later maybe. The problem is, the way it creates the bust doesn't mean it would need to be that way.

>>32788 These are sooooo good. Is it possible to get a side view. I wonder how the eyes look not face center. I need to watch that Alita movie and pay attention. I have it.

>>32799 > it possible to get a side view. I don't think I can make the side view of a specific picture, maybe if I don't use a random seed next time, keep the same and change the prompt to side view. I might try that out later. > I wonder how the eyes look not face center. Most likely because of the phone, idk. > I need to watch that Alita movie and pay attention. I have it. You never watched it?!

>>32803 Wow! Those last two definitely have an excellent Alita-like vibe going. GG, Anon. Cheers. :^)

>>32803 I made some from the side, but no comparison with the same seed than the front view yet. Also tried to add some shoulders and chest.

>>32806 Color switch...

>>32807 When a good picture is found, its important to copy the seed and untick the "random seed" option. Then increase or decrease it by one at the time. Copying helps if the tab reloads. That said, with limited GPU access it's important to not forget to change the seed, otherwise it will create the same picture again. This can also be useful, if you want to change something. I could've removed the botched freckles in picrel 3 in >>32807 and maybe improved the eyes by using the same seed but changing the prompt a little bit. If you keep changing the seed and keeping the prompt at some point it might still switch to less realistic pictures, which might or might not be good. Though I also played with the CFG scale value.

>>32803 >You never watched it?! Yes, but it was a while ago and I didn't really study it as a template or paying close attention.

>>32808 Thanks, much appreciation. These don't really look bad in terms of bug eye. In fact the look great. I don't know if it's these particular models or if I'm just getting used to the look. Visually judging things in an artistic, technical way is not easy. Otherwise all of us would be great artist, but we're not.

If it's not a huge secret, what prompts are you using to get this great work, and are you using some other picture as a starting point and feeding it to the AI? Is it even possible to feed it a stating point picture? This is really great stuff. Now if you could get several shots of the same model from any positions there's software that will build a 3D model from that. So you could have some really good stuff chaining different told together, I think, maybe.

>>32817 Well, I had different variants, e.g. >realistic picture of an young adult woman, doll-like but otherwise realistic looking eyes with a realistic eyeball, very realistic lightblonde hair, realistic eye brows out of human hair, bluegreen eye color and realistic looking skin, very smooth lips without folds, small chin, high cheekbones It's more or less the same than I posted already here >>32673. This resulted in the portraits. I used a somewhat increased CFG scale. I think I let the other values the same, until later. I also went mostly with the realism Lora, and with Boreal a few times. > both eyes in the same color, very realistic looking textures skin, Was also added. I think anything with the tongue doesn't work anymore. I seem to be the one trying this on every site until they block it. I got some photos I won't post, but not many. For the ones with the chest: >realistic picture of an young adult woman, showing the face and upper body, wearing a shoulder-less blue dress, doll-like but otherwise realistic looking eyes with a realistic eyeball, very realistic whiteblonde hair, realistic eye brows out of human hair, very blue eye color, both eyes in the same color, very realistic looking skin, very smooth lips without folds, small chin, high cheekbones Of course, colors can be changed. I also went with swimsuit instead of a dress later. Picrels were made with >realistic picture of an young woman, taken from a isometric perspective and higher position, showing the face and upper body, wearing a shoulder-less pink swimsuit, doll-like but otherwise realistic looking eyes with a realistic eyeball, very realistic light red hair in two tails, realistic eye brows out of human hair, grey blue eye color, both eyes in the same color, very realistic looking textures skin, very smooth lips without folds, small chin, high cheekbones, - long neck, realistic neck length Still very often good looking, but I have a hard time to get it to make realistic photos, it rather goes with doll-like plastic looks. Not in all cases, but much more often and very much so. Not sure if they changed the rules, or if it is because of the color change towards pink dress. Now it also ignores my demand for a realistic neck length.

These are really nice, NoidoDev. About the (pretty severe) disconnect between the exported meshes and the final rendered images with this tool, I have a theory: Among the many types of maps you can apply to 3D polygonal meshes (texture, normal, glint, etc.), one of them (and an important one for vidya especially) is bump map. It's possible they have an algorithm to adjust-for the large deformations of the underlying mesh (whether there intentionally or not) to bring the render up to a high quality during the render pipeline. Unfortunately, if this is in fact what's going on -- or close to -- then we'll be unlikely to use this tool as a source for robowaifu head modelling. It would take some pretty sophisticated engineering to 'reverse' this process, and apply the bump map offsets back onto the base mesh to fix it. It would take roughly the equivalent of a render engine in complexity, I'd estimate. Well above my current knowledge, certainly. <---> Regardless, these are still some very nice busts Anon. Quite inspirational for Anons, I'm sure. Cheers. :^) >=== -sp, fmt, prose edit

>>32818 Robot bodies also work. Replace the swimsuit line with something like > the upper body has some robotic shell elements in dark blue soft plastic If you mess it up somehow, then it switches to shiny white shells. This seems to be the standard. I wanted something like TPU or silicone but still a shell with different elements. I also wanted something like white hair with a light blue hue, but didn't get it. This here works for light blue: > extremely light blue hair The reason why I got even into this more general modeling was a bit of distraction, including making videos with Hedra. Unfortunately, most pictures I tried today failed. Only one worked, and this is one with nothing else in the picture than the face. I guess they messed up their model somehow. It got better in some ways, but worse in others. This is just me playing around anyways and I don't get many views, so I might stop it after a while. Signing up to Kling AI also didn't work, and I don't have the nerve to deal with such nonsense. >>32821 Thanks, good to know. I have other plans anyways. I might to use some of those busts as patterns for modelling something. Maybe have a start model to go from. I also have another idea which I don't want to share yet.

>>32818 Thanks for the pics and the generation explanations.

>>32839 The robowaifu police will keep you safe. On a more serious note, I was testing a softer robot shell: > the upper body has some robotic shell elements in dark blue soft plastic broken up into sections, the shell is not a suit or dress, Important difference is, I switched to the "Boreal" Lora away from the "Realism" one. This one understands most times somewhat what I mean by soft plastic, though it still surprises me, the other one didn't get it at all just went with hard shells. I also removed the small chest option I had before, which I had since when I used to make pics of swimsuits I got a loot of quite big boobs. Which isn't bad, but I didn't ask for it and it was a but much. When switching to robot shells, I had to remove it, otherwise the boobs were very small or none existent (last picrel). Oh, and especially picrel one is another piece of gold, IMO.