/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

The Mongolian Tugrik has recovered its original value thanks to clever trade agreements facilitated by Ukhnaagiin Khürelsükh throat singing at Xi Jinping.

The website will stay a LynxChan instance. Thanks for flying AlogSpace! --robi

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB


(used to delete files and postings)

Knowing more than 100% of what we knew the moment before! Go beyond! Plus! Ultra!

Robo Face Development Robowaifu Technician 09/09/2019 (Mon) 02:08:16 No.9
This thread is dedicated to the study, design, and engineering of a cute face for robots.
>>4549 >Spoiler: For faces sculpting seems to be better. What I'm wondering is, if you can take a bunch of professionally-sculpted face meshes, run them through a number of parameterization algorithms, and then come up with a few sound principles for beauty and appeal (Nordic women's facial forms say) that are basically automatable using a GAN-like approach after the parametric analysis has been done.
>>8230 >that are basically automatable Just to clarify, I mean that the design generation itself is automatable, not the topic of automating the robowaifu face after manufacture.
>>8231 I'm sure something like such a generator will be available at some point. However, I was rather thinking about one that would take in photos of a lot of pretty faces and then create new ones, then make a mesh out of the selected one. Two separate steps, which I think exist already on their own. Or a bunch of photos of a e.g. an actress would go in, taken from different angles, and then it would make a model. With enough models it could get better to put out something pretty. However, I was thinking that a user would give it the name of some public figures (actress or model) and it would generate some examples which would be close but not exactly the same. From those the user could then choose the preferred ones. Also making it more anime-like (Alita style) looking. I don't really think that some highly specific facial features of nordic women exist and that it make sense to go after that, though. Just make the eyes blue or green and the hair something between blonde and red or light brown, and maybe add some freckles. To fit in the huge eyes, the face would also have to change anyways... Also, if it goes after the pattern of an actress then it should put out what you want, if this includes something nordic then it should be in there. One of my ideas in that area was, that someone (group, company or person) could pay low-wage sculptors in poor countries to make models from a pool of pretty women (actresses and models). Then this would be a pool which could be used directly, or to train such generators on the creation of new but similar face meshes. After all, don't forget we need the skulls for the faces as well, so they fit onto it, and then a way to animate it with ease.
>>8232 You're correct that tools exist to do facial feature extraction to generate facial meshes (and other kinds) using multi-camera setups. An example of this technique (with a different kind of focus) is here >>1088 . But afaik, they are still fully proprietary and highly expensive. Addmittedly I haven't looked into this area for a year or two so maybe good opensauce systems exist now, let us hope so. Certainly the all-in-one '3D' cameras are more numerous to choose from now. And as you suggest, collections of images can potentially stand in proxy for such a method, though the 'registration' part of the process is both tedious and error-prone. >To fit in the huge eyes, the face would also have to change anyways Fair enough. As much as I like Alita, I think Rodriguez ( >>4502 ) went just a touch overboard with the design. But really, the critique is only because the rendering otherwise is so good and so realistic that it triggers a bit of a 'wut' in me (and others). As a counter-example, here's work by an artist that I think has found a near-perfect balance between kawaii-eyes and facial realism, though in an artistic style. > One pretty famous example of your pool of artists idea that has actually been carried by a very multi-talented guy is Ricky Ma's avatar effort >>153 . The very fact he created a small furor among women's advocacy groups as a 'creepy stalker' shows just how good a job he's done with it. Hopefully /robowaifu/ can manage to produce many, many examples that will do just as well (or even better)! :^) >--- -Update: Welp, I've attempted six times now to post this pic for you, obviously nuJulay won't cooperate atm. I'll try to do it again for you ITT later on Anon. The artist is named Ivant Alavera.
>>8236 >Ivan Talavera*
>>8240 >>8241 Ah, thanks. Cute, but I'm also fine with Alita. I think the bigger eyes might help if one looks rather young otherwise, to make sure she doesn't look human. >>8236 In software for users to edit I found this: Gradient Mesh Illustrator - https://youtu.be/JEJHk9VRAEQ and similar stuff. Gradient Mesh seems to be the term to look for. Also VoluMax might help: https://youtu.be/4XdoN2-8Dg8 However, my point is raher that we don't need to copy some specific face anyways, and this here is from 3 years ago, and a photo seems to be enough: https://youtu.be/u9UUWqVquXo - we only need to have it in a way that we could print molds from that. There's more on their site: http://www.hao-li.com/Hao_Li/Hao_Li_-_publications.html
Anyone seen this video? It does a decent breakdown of how unscientific most memes of the "uncanny valley" actually are. It may also help to clarify goals regarding robowaifu function and design (particularly of the head/face). https://www.youtube.com/watch?v=LKJBND_IRdI
>>8261 Yes, you probably have that from this site here, because I posted it (and on cuckchan and on other occasions).
>>8266 Seems the conclusion is that all robots are designed to perform a specific task or small set of tasks. Robowaifus are, for the most part designed to elicit a positive emotional response from the user. So I shouldn't try to make a robot in the image of a human that can carry out all of the tasks a human can, because I will just fail at both. Instead I think I'll just focus more on making a cute robowaifu who fulfills the function of reducing loneliness, rather than attempting to make one that can walk or perform complex motor actions like play sports etc. We already have some pretty high-quality chatbot software in the form of GPT-3, which I should be able to link to my current text-to- speech program- so I just need to complete a reasonable looking robowaifu body now.
>>8267 Being a cute waifu is the priority, but then improving her within these constraints. That's they way. I guess it will be rather so, that the more skilled ones will be more expensive.
Open file (268.63 KB 1240x897 MjcxNzYxOQ.jpeg)
Hey for those of us going for the human look, the use of prosthetics are always a good possibility. For the face, actual dental replacements (dentures?) could provide highly realistic looking teeth for a waifu's bright, sunshiney smile. :^)
Open file (191.52 KB 802x1202 summer-glau_02.jpg)
>>8326 Correct, they had this idea mentioned in the original board on 8chan. I never looked into it so far, where these are available, what they cost and where to get them. There must be some sources for training of dentists or something. They seem to have some kind of dolls with fake teeth to train on. Not sure how hard they are, though. I hope we get them made out of ceramics in some standardized sizes and don't need to build them on our own as well.
>>8331 Yea good thinking Anon. I bet we can source them somewhere on the cheap. Remember the mouth needs to be kept sanitized just like w/ humans so the source needs to be reputable. Remember your robowaifu will probably need to kiss you lots to stay happy! :^)
While not strictly RoboFace development per se, until we have a dedicated MOCAP thread this might be a good spot for this. I indirectly discovered this project today after looking into SingularityNet via Anon's post. >>8475 . It's a tool that finds facial landmarks in video. Helpful for thing like facial retargeting, etc. https://github.com/singnet/face-services
Open file (1.30 MB 2731x4096 IMG_20210325_183338.jpg)
Open file (585.69 KB 2731x4096 IMG_20210325_183326.jpg)
Open file (1.01 MB 2731x4096 IMG_20210325_183343.jpg)
This is what is possible today, though lightning might be relevant for the look. Let's see how they'll look after being shipped, and some customers take fotos and report on it. These are the Alita busts from Queens Studio. I already mentioned them here: >>8194
>>9260 Yep that's nice Anon, thanks for the updates.
So, things are moving forward here. New network creates toonified faces out of real or made up ones, also allowes mixing of two input pictures. >Our ReStyle scheme leverages the progress of recent StyleGAN encoders for inverting real images and introduces an iterative refinement mechansim that gradually converges to a more accurate inversion of real images in a self-correcting manner. https://yuval-alaluf.github.io/restyle-encoder/
>>10461 Oh, video: --write-sub --write-description https://youtu.be/9RzCZZBjlxM
>>10463 Thanks very much for taking the extra time to give a more full youtube-dl command to use Anon. Getting and keeping the description and subs will be important to anyone keeping personal archive of YT videos, once cancel-culture Marxism literally deletes anything/everything that could possibly have any bearing whatsoever on either robowaifu creation, or anything else that could possibly help men. Since the Lynxchan software adds an '[Embed]' into the text of the command, I always put such a command here on /robowaifu/ inside codeblocks, since the CSS here disables this embed tag. youtube-dl --write-description --write-auto-sub --sub-lang="en" https://youtu.be/9RzCZZBjlxM
>>10461 >>10463 That's really cool Anon. He's humorous to listen to as well, his enthusiasm is great.
That screen one looks nice for nanny.
https://www.thingiverse.com/thing:4865223 https://www.youtube.com/watch?v=8_wkbLL0fqM LED Matrix behind tinted plastic, cheap, easy, customizable
>>12938 Thanks, this might be something fitting very well into the basic idea of the board: Making affordable robowaifus, which don't need to look like humans but can be a bit more on the robot side.
>related crosspost (>>13020)
Open file (47.85 KB 600x414 uploads.jpg)
>>12938 That's like Rina Chans board. She's am autistic that uses a board to convey her emotions because making facial expressions is hard for her. Her board could be a really cute face for a robowaifu.
>>13560 this will be a thing in the 5-10 years all those kids growing up around masktards are going to be incapable of expressing emotions
>>13560 This could probably get quite expressive with a high enough resolution.
Open file (108.85 KB 335x640 mace_griffin_acolyte.jpg)
>>13560 >>13563 >>13565 I don't know. It just reminds me of the cultist NPCs from the game Mace Griffin. Using a display seems like a trade-off for using a real tangible face, if you want more face customization, but a emoticon-like face just seems like a really bad trade-off.
Simplified non-human faces go a long way towards bypassing the uncanny valley
>>12950 >Making affordable robowaifus, which don't need to look like humans but can be a bit more on the robot side. Agreed. >>14870 I don't personally find that particular face appealing, but I think in large part you're correct Anon. At the very least, I'd suggest we all seek for ways to maximize the expressive potential of, while minimizing the underlying complexities of, our robowaifu's face systems and structures.
Open file (42.60 KB 720x540 MaidroidMiao.jpg)
Open file (113.77 KB 750x1000 Sinobu.jpg)
>>14935 Could you provide examples of faces that work in 3D at human scale? Most figures are cute at figure scale but, they look uncanny at human scale.
>>14955 The life sized Rem statue is the best I've seen with a full anime aesthetic.
Open file (527.88 KB 2048x1364 E0SSS1JUUAEoWCw.jpg)
Open file (477.94 KB 1152x2048 Ezfhy32VIAE6pIS.jpeg)
Open file (565.04 KB 1364x2048 EzpxOz7VIAMKsjY.jpg)
Open file (1.38 MB 1920x2560 EWSMrC3UwAE-KkL2.jpg)
Open file (605.49 KB 1707x2560 1566424488789.jpg)
>>14955 Anime is inherently doll-like so at human scale it's bound to feel unnatural when you go to kiss her on the cheek and her eyes are twice the size of yours. Even most weebs will probably prefer something semi-realistic. It's really a matter of personal preference though. Some find anime dolls creepy and uncanny but I completely adore them. People will have to come to terms with robowaifus being animated machines and develop new aesthetics that feel comfortable to live with.
>>15285 Also something to consider is faces will feel very different once they start moving around. SEER's cable eyebrows aren't human-like but they add a lot of expression.
Open file (92.82 KB 988x1478 Kyoko.jpg)
>>15280 That Rem is gorgeous Her roundness and cat like features do wonders. >>15285 Good point about the eyes. Human eyes are approximately an inch in diameter, so that could be a dimension to keep in mind. The ones with smaller noses and less human-like features are the cutest. There seems to be some middle ground between cats and girls that's best. Also, fangs are amazing! >>15287 This is an important point. Looking good in motion is vitally important.
>>15280 Wonderful figure Anon, thanks for sharing it. >>15285 >>15287 Good points. >>15288 A cute. I think we can all take pointers from the dolls/figurines world for design aesthetics.
Open file (141.05 KB 1024x768 humaneyes.jpg)
Open file (296.59 KB 711x336 roboeyes.png)
Main aesthetic issues I see with regard to making realistic robot eyes: -Eyelashes often get punched in all crazy. This robot's eyelashes (specifically on the lower half) are way too sparse and erratic, and don't extend to the outer corners of its eyes. -Eyes do not sit right within the face. Notice how there's this gaping hole in the robot where the real person has muscle. Realistic eyes need muscle surrounding them like realistic teeth need gums. I understand the eyes need a way to move around without lubrication, though, so this is a real issue. -Again, this probably has to do with eye movement, but the eyelid creases are too deep. I've never seen anyone who has a weird overhang like the one above the left eye and there's something about the way the upper skin hangs above the crease that makes it look old and "crumbly".
Open file (284.34 KB 634x424 ClipboardImage.png)
Open file (232.91 KB 474x355 ClipboardImage.png)
>>15997 The eye crease is unique to Engineered Arts robots. Take a look at Asuna's eyes, they are very cute!
>>15997 I'm quite sure the eyelashes are just a matter of how much work to put in, and about finding the right technique. The gap problem is something I realized myself, but this might differ based on the construction. Also, silicone rubber is actually very flexible and wouldn't resist that much. This might just not being used because these are all just prototypes.
>>15997 >>15998 >>16001 My $.02 on the matter of hyper-realism for eyes is quite simple. In a word: don't. As discussed briefly here (>>15962, ...), and frankly, across this board for years now, is that we aren't in any position (not just us, I mean mankind generally) to 'escape' the Uncanny Valley for hyper-real, general-use Gynoids/Androids/Mammaloids anytime "Real Soon Now"(tm). We're simply too well-tuned by God as homo sapiens sapiens to be fooled easily, just yet. And 'human' eyes in particular are a relatively small part of the body that a great proportion of our brains & souls are tuned specifically for. I don't want to discourage anyone here from proving me wrong (far from it, I eagerly look forward to that day! :^), but tbh it's a fools errand IMO to think we'll manage such right out of the gate. Therefore, I agree personally with the general consensus on /robowaifu/ thus far (indeed our tagline): >"Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality." At least the cute 'anime' bit at a minimum for now. Any other approach in the near-term will be a disappointing fall into that infamous valley IMO. Thoughts? >=== -minor spelling edit -minor grammar edit -minor prose edit
Edited last time by Chobitsu on 04/25/2022 (Mon) 09:22:35.
>>16004 >matter of hyper-realism for eyes is quite simple. In a word: don't. Devs should try, then we'll see. Filling the gaps in the holes isn't really ruining anything. >escape' the Uncanny Valley >fall into that infamous valley I always like to remind people of this video: https://www.youtube.com/watch?v=LKJBND_IRdI Just don't make it look creepy nor ugly.
>>16008 This video also addresses the history of "uncanny valley" and the fear of human-like dolls: >Why Are We Afraid Of Dolls? https://www.youtube.com/watch?v=nHzwMVpiLD4 It's basically a attractive idea for creators to use dolls for horror, because they were meant to be fun (similar thing with clowns). The problem is, too man people let their media consumption shape their world view. Some might get traumatized by watching horror movies while being to young. >Pediophobia: The Fear of Dolls https://www.verywellmind.com/fear-of-dolls-2671875 https://www.healthline.com/health/mental-health/pediophobia Around 9% (of Americans) seem to have it, more women than men. Exposure seems to be a important part of each therapy. >The outlook is very good for people with pediophobia who seek counseling for their phobia. To improve the outlook, a person with pediophobia needs to be fully committed to their treatment plan.
>>16008 My $0.02 is that the Uncanny Valley theory (at least the media's interpretation of it) isn't actually 100% accurate. Specifically the notion that there's a "no zone" somewhere between inanimate-looking and totally human-looking that everything within is creepy. Rather, I think that as intended human-likeness increases, the potential appeal increases in a linear fashion. However, the potential creepiness ALSO increases linearly. An extremely realistic robot has the highest potential for human appeal, but is also the hardest to execute well and will be the most off-putting if not well executed. Uncanny Risk vs. Reward, if you will. Consider something like the CG in a recent Final Fantasy game. The characters are ALMOST human but not quite - according to the Uncanny Valley, they should be extremely unsettling. Yet that's not the case, I'd imagine most people would rate them somewhere between "mildly creepy" and "highly appealing". (reminder: we're just talking about visuals here) Now contrast that with the IEEE Creepy Robots poll (https://robots.ieee.org/robots/?t=rankings-creepiest-robots) and you will notice that none of the top 10 are nearly as human like as my aforementioned Final Fantasy example, but I imagine most would consider any one of them far, far creepier. In fact, the human likeness varies a lot between the top ten - there are plenty of examples of robots people would consider less unsettling than the top 10 in this list despite them being more human-like. Compare Pepper and Kobian, for example. Or Cutieroid and CB-2. The biggest implication of this is that the best way to fix a creepy robot design is not to "move it left/right of the Valley" by changing it's human likeness, but rather to improve the execution of the design itself. While making a robot less human like can resolve the creepiness issue, it does so at the cost of appeal.
>>15997 Though you are correct on your aesthetic reasoning, realistic faces cannot be made by man. There are numerous flaws you did not mention which would still fall into the uncanny valley. >>16004 Chobitsu is correct, going for a demi-human/anime inspired aesthetic is much more attainable. >>16008 One of my favorite videos. Though it is correct to say the uncanny valley is not real, it's a useful term to describe the general unease felt by these creepy machines. I do think good can come from researching solutions to the uncanny problem but, that's best left to corporations like Disney. >>16012 I think "cuteness uber alles" is the best way of thinking about it. Almost human can be quite cute indeed, it's just easy for almost human to end up creepy like Sophia. This is why I think sidestepping into the demi-human route is good. I also want to clang monster girls though so, I'm heavily biased. Picrel has big fins, a flat face, and spiky red eyelashes without eyebrows. Almost human but, not quite.
This is a great conversation going r/n ITT, /robowaifu/. I hope we can keep it going for a while, and I'd like to link to it in our mini-FAQ in the /meta's
>>16008 My 2 cents is that the uncanny valley dont exist and dont matter. This is robotics not a beauty competition, so we should be caring about the technical aspects, this is why i think robots like Sophia are inspiring. This is good since there are actual challenges that need to solved like budget, time, hardware, software and more
Open file (273.40 KB 1440x1800 1652407800269.jpg)
Open file (153.27 KB 1080x1080 1652403268055.jpg)
Open file (335.35 KB 1440x1800 1652403080234.jpg)
>>15285 After more research, I've come to the conclusion that anime doll faces are the only way to make her cute and simple. All of these could be printed on an SLA printer relatively easily and the eyes would also be relatively easy to make. The hair would somewhat of a problem but, adding non-human ears like cat or wold ears would help. Go for the demi-human route.
Open file (104.56 KB 800x1200 1607113054993.jpg)
>>16246 >After more research, I've come to the conclusion that anime doll faces are the only way to make her cute and simple. I've suggested all along that the doll industries have much to offer us in our quest for robowaifus. This isn't a new conception either.

Report/Delete/Moderation Forms