/robowaifu/ - DIY Robot Wives

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

Happy New Year!

The recovered files have been restored.

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


“Stubbornly persist, and you will find that the limits of your stubbornness go well beyond the stubbornness of your limits.” -t. Robert Brault


DCC Design Tools & Training Robowaifu Technician 09/18/2019 (Wed) 11:42:32 No.415
Creating robowaifus requires lots and lots of design and artistry. It's not all just nerd stuff you know Anon! :^) ITT: Add any techniques, tips, tricks, or content links for any Digital Content Creation (DCC) tools and training to use when making robowaifus. >note: This isn't the 'Post Your Robowaifu-related OC Arts & Designs General' thread. We'll make one of those later perhaps. >--- I spent some time playing with the program Makehuman and I'll say I wasn't impressed. It's not possible to make anime real using Makehuman, in fact the options (for example eye size) are somewhat limited. But there's good news, don't worry! The creator of Makehuman went on to create a blender plugin called ManuelBastioniLAB which is much better (and has a much more humorous name). The plugin is easy to install and use (especially if you have a little blender experience). There are three different anime female defaults that are all pretty cute. (Pictured is a slightly modified Anime Female 2.) There are sliders to modify everything from finger length to eye position to boob size. It even has a posable skeleton. Unfortunately the anime models don't have inverse kinematic skeletons which are much easier to pose. Going forward I'm probably going to use MasturBationLABManuelBastioniLAB as the starting point for my designs. --- Saving everything with yt-dlp (>>16357, >>12247) >=== -re-purpose OP for DCC consolidation -add yt-dlp crosslink
Edited last time by Chobitsu on 05/31/2023 (Wed) 18:55:38.
After posting this above, I found this: >AnchorSCAD is a Python 3D modeling API for OpenSCAD. This document covers the minimal set of concepts you need to understand in order to build shapes using AnchorSCAD. https://github.com/owebeeone/anchorscad
>Learn how to control #servos based on a #Blender animation with a dedicated Add-on I created. This technique helped me to efficiently animate my DIY #animatronics and is based on a process I picked up while watching one of Disney’s behind the scenes video. https://youtu.be/yeJxMaNQAzg >Export your Blender animation to servo position values. They can be used with an Arduino compatible micro controller to move PWM-driven servos according to your animation. https://github.com/timhendriks93/blender-servo-animation
>>31389 Neat! Yes, using Blender as a kind of development/staging/simulation environment for robowaifu design efforts seems to me a very good idea (for a long time now). This handy plug-in can be just the thing for quick motion prototyping! :^)
>>31502 I think that's the plugin I use for SPUD. I had to modify the export script though because it exported in some crazy, super long format with lots of { and }. If I remember right the export lists a servo, then all the positions of that one servo eg servo A postions for frames 1-24, servo B positions for frames 1-24... Its easier just to export each servo position as an item in a list, 1 list for each frame on 1 line... if that makes any sense XD
>>31517 Oh right, I do remember that now Mechnomancer, my apologies. I'd be interested to learn detailed evaluations from you about your experiences with it. What works well, what doesn't, things like that. --- BTW, while I've got you 'on the line', I'd like to point out my recent post in the MaidCom thread : (>>31482) . I'd like every Anon's view on this matter of developing baseline robowaifu hands we can all use/extend for our own projects -- but particularly from you rn, since SPUD appears to be in need of some dextrous hands. I'd suggest we all earnestly consider Will Cogley's Bionic Hands project [1]. Also, I suppose the board should conduct this group-conversation within our Robo Hands thread : (>>4577) ? If you agree please make a new response there. Cheers. :^) --- 1. https://www.youtube.com/watch?v=8w88HwbYPWI&list=PL6AJl-knQFCGJakVBP1_3EXqJyAS0qN8k&index=18 >=== -fmt edit
Edited last time by Chobitsu on 06/09/2024 (Sun) 18:26:01.
>>31522 That's ok. SPUD actually does have hands (a prefab set off Spamazon), I just don't have them installed right now (will probably buy a second pair of rubber hands, gut them and try to stuff the robohands in there). You can see a little peek of them in this post which was my first attempt at servo frame animation >>31324 For the plugin you need to know how to rig stuff in Blender and keep track of which bones are labeled what, knowing how to translate the servo pulses into degrees and putting limits on the bones. Not particularly difficult if you've been doing 3d modeling for 20 years (god has it been that long already?) like me but new folks could find it challenging.
>>31537 Good luck with those development efforts. Looking forward to seeing them working. >just master Blender first, bro... Lol. OK, will do Anon! :D
>>31517 >I think that's the plugin I use for SPUD. That's almost certain where I got the link from, but I didn't have the reference anymore when I posted it. Good to know, thanks. I started trying to go through some lists of videos I have and post them here, but in case if something came from another thread or the Discord I won't remember that and can't search the whole board right now.
Going to leave this ITT, till I figure out if it should be moved. --- Quite a good demonstration of all 12 Principles (of animation), done by a professional: https://www.youtube.com/watch?v=gyGEwyWLbHk
>>24479 Related: >>32638
Open file (74.94 KB 489x370 Meshy_3D_models.png)
>>32670 and >>32638 Here's my profile on Meshy: https://www.meshy.ai/@NoidoDev - where I can publish the hard surface models, but most of the ones looking good are soft ones and they can't be published. I downloaded them, though. I also found a 3D robot girl on Meshy, but only pro users can download it, it's on my "collected" page. But I used the prompt to make my own and got a decent but not as clean result (picrels). I also saw a tutorial where people were able to repair models using AI on that website, but I think it was pro user as well. It works to animate models and then using Blender to make it better. I also tried text to 3D on the Meshy website a few times more, for a Waifu sculpt and I didn't get a good result. Mega Folder for 3D bust and AI 3D models: https://mega.nz/folder/v7Z1DBLA#xsx9KqZHqr_eMJ-MNNjejA
>>32735 Thanks, NoidoDev! Great job.
Open file (160.08 KB 511x641 Bust_Imports_002.png)
Open file (162.42 KB 511x641 Bust_Imports_001.png)
>>32735 So, I imported some of the head busts into OpenSCAD. Picrel is how they look like. Not great, but I might do some experimenting with it. One idea is to mirror each side of a splitted face and combine all of it. The other idea is, to just use these as patterns to remodel them better. I also could try briefly in MeshLab, if it can do any automagic. That aside, I want to note that I would only want the faces of these heads anyways. I think the head needs to be done with OpenSCAD elements. I will post such more detailed testing in the current prototype thread.
>>32770 Yeah, my results are the same in Blender, NoidoDev. Much like the texture obfuscation that anons were dealing with a couple years or so ago, I expect this tool is pulling some kind of malarkey w/ the mesh data perhaps? Just keep moving forward, Anon. We'll arrive at good results in the end I'm sure. Cheers! :^)
>>32772 >texture obfuscation that anons were dealing with Not sure what you mean. There was an issue which was more like something to avoid "stealing" the models and use them for something else. Here it's more like the models looks better in the preview. The textures made the problems look less bad on the website, but I don't think the models have been made bad to avoid further use. The holes in the head for example were visible. This is just from a bad translation from 2D to 3D. CSM AI did it better: >>24484 >>32770 Anyways, I did some of these tests right away with one that I picked on a whim (bottom right). First picrel is each half mirrored, second one is the same but showing the halfs, the third is then blended the two heads into one with some adjustments.
I've been distracted by other work, and didn't realize Blender 4.2 LTS came out a couple months ago. It's remarkably impressive. Check it out! :^) https://www.youtube.com/watch?v=-mTk1hQLlTY
Open file (64.13 KB 644x403 image2edit.png)
Open file (143.31 KB 511x641 HeightMap01_001.png)
>>32777 So, I'm working on trying out getting the facial features into a height map. Once again, this will not lead to anything useful soon, but might be interesting long term. For automatic face generation. Also, my tests are still in early stages. - https://en.wikipedia.org/wiki/Heightmap - https://en.wikibooks.org/wiki/OpenSCAD_User_Manual/Other_Language_Features#surface - https://cloudcompare.org/forum/viewtopic.php?p=11484 - https://www.cloudcompare.org/doc/wiki/index.php?title=Rasterize Related: >>32793
>>33072 Very cool! Glad to see you researching this topic, NoidoDev. Good luck! Cheers. :^)
The latest Beginner's Blender Donut Modelling Demo : https://www.youtube.com/watch?v=B0J27sf9N1Y
>>23849 > List of other 3D programmatic modelers This really is a great list. Thanks. I'll save in my robowaifu text file
I've been looking at making molds but trying to make it with the cheapest stuff I can find and thought of paper mache. I want to do some boat stuff mods for fiberglass. The traditional way to make forms with wood and fiberglass is super expensive. One way to get large forms with curves is a technique called "fabric formwork". Using that search term (look at images for links) you can find these super awesome fabric forms they use to make kick ass concrete stuff. Buildings, bridges, all sorts of stuff. They use that fabric that is used for silt control on construction sites. Make a form that drapes and fill with concrete. The water drains out leaving a super strong, and very curvy good looking concrete item when hard. https://www.taylorfrancis.com/books/edit/10.4324/9781315675022/fabric-formwork-book-mark-west I was thinking of large boat structures, making forms with this draping technique, then using these forms to vacuum form fiberglass. The same could be used for waifus. So I look around for paper mache to make cheap forms and found the mother of all paper mache sites run by this nice Grandma, or so she says. Here's a link to her 40, paper mache recipes. All sorts. She has a nice clay type that can be used to model like modeling clay. I suspect that you could model waifus a lot faster with an actual material like this than using a computer and then printing it out. And super, super cheap compared to the tech way. Here's a link to her recipes list. Some are water proof some lower cost, some stiffer, all sorts with videos. https://www.ultimatepapermache.com/recipes-for-paper-mache If you scroll down this page she gives some good references to sculpting books and resources. She also has some good sculpting ideas. From that page and others of hers you will find a number of methods for sculpting. Once you have these sculpted you can make molds for them. https://www.ultimatepapermache.com/paper-mache-armature-patterns Here's another site on Tor where you can find books on sculpting, human anatomy and even books on skeletons. This one site has enough data in books to cover most any referencing you need to make models of, skeleton, muscles, etc. http://libraryfyuybp7oyidyya3ah5xvwgyx6weauoini7zyz555litmmumad.onion/ Search terms on site Human anatomy Trail Guide to the Body Skeletal Muscles Biomechanics Biomechanical Anatomy Sculpture Human Gait Human Body Life Drawing Skeleton Musculoskeletal Models Running Mechanics "discover the story of human and animal skeletons" I thought this one was nice "QuickStudy - Muscular Origins and Insertions " and for lite reading "Pedophilia & Empire Satan, Sodomy..."
>>33119 Great stuff. Helpful.
>>23844 >You are correct, I'm working on spherical actuators that should be scalable for use as a universal robotics joint. Just curious what the outcome of your research was, Anon. This would still be a tremendous benefit to us all to have 4 of these for the shoulders & hips.
>>33321 Added complexity and frankly, just simple pan tilt mechanisms ended up being better.
>>33318 Great stuff, Grommet. Thank you kindly! Cheers. :^) >>33322 Understood. We'll arrive at good solutions for everything in the end. Let's all do our best -- Gambatte!! :^)
>Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation [1][2][3] >abstract: >We present Hunyuan3D 2.0, an advanced large-scale 3D synthesis system for generating high-resolution textured 3D assets. This system includes two foundation components: a large-scale shape generation model -- Hunyuan3D-DiT, and a large-scale texture synthesis model -- Hunyuan3D-Paint. The shape generative model, built on a scalable flow-based diffusion transformer, aims to create geometry that properly aligns with a given condition image, laying a solid foundation for downstream applications. The texture synthesis model, benefiting from strong geometric and diffusion priors, produces high-resolution and vibrant texture maps for either generated or hand-crafted meshes. Furthermore, we build Hunyuan3D-Studio -- a versatile, user-friendly production platform that simplifies the re-creation process of 3D assets. It allows both professional and amateur users to manipulate or even animate their meshes efficiently. We systematically evaluate our models, showing that Hunyuan3D 2.0 outperforms previous state-of-the-art models, including the open-source models and closed-source models in geometry details, condition alignment, texture quality, and etc. Hunyuan3D 2.0 is publicly released in order to fill the gaps in the open-source 3D community for large-scale foundation generative models. The code and pre-trained weights of our models are available at: this https URL --- 1. https://arxiv.org/abs/2501.12202 2. https://github.com/Tencent/Hunyuan3D-2 3. https://huggingface.co/tencent/Hunyuan3D-2
>>36183 >Hunyuan3D-DiT This looks great but it's not clear to me if you can input many different pictures taken from different angles and then have it stitch those into a 3D object OR if it just uses one picture. How it would do this, I have no clue but they sort of show one picture and then many pictures form several sides. It's confusing. "If" it does many to 3D, then I really need this but of course don't have the video card to run it. Sigh...but sooner or later some one will find a way to run it on a CPU. So it's all good.
Open file (176.21 KB 676x709 Hunyuan3D_Test01.png)
Open file (184.63 KB 676x709 Hunyuan3D_Test02.png)
Open file (204.58 KB 676x709 Hunyuan3D_Test03.png)
Open file (195.16 KB 676x709 Hunyuan3D_Test04.png)
>>36183 Thanks, I already saw that somewhere and thought about posting it here. >HuggingFace It works fast and gives decent results. I will post some here, but if I try out more further results will go to the AI art thread. I will make a crosslink anyways. Here my post on meshy.ai and why I wanted that: >>32653 - Bust to 3D model of a head. The results from my test so far are okay and better than the ones from Meshy, but it has the old problem of making the main sphere of the head too big. This can maybe be mitigated later in a 3D program, though. It also creates some other errors, though, and interprets the face too much and changes it. It especially seems to make the mouth into something sad looking or sulking, idk. Anyways, it's certainly progress. Then again, unfortunately the better model is somewhat closed source (but free?). Also, downloading the "White Mesh" on HuggingFace didn't always work right now, the button didn't respond. So, I don't have all files right now. But some, if someone needs them I might upload them, on Catbox and sooner or later on my Mega account anyways. >>36185 It being on HuggingFace means you can go to a space and try it out. Or you make an account, give them money and you might be able to make your own "space" with the model running on their servers, while you're only paying for uptime. This is mostly necessary if there is no free demo space or you want to use it more often than HF allows it for free.
>>36185 >This looks great but it's not clear to me if you can input many different pictures taken from different angles and then have it stitch those into a 3D object OR if it just uses one picture. It looks to me that this is good enough for at least one of us here to find out, Grommet. So just keep an eye out. And in the meantime, shop around and pick up at least an old-a*rse GPU, make sense? >>36287 Wow! That's really impressive, NoidoDev. Thanks for sharing these. I hope you share more in the other thread too. Are you planning to take advantage of the texturing capabilities of this model as well?
>>36294 I think this will be useful in some time, when we want to make individual faces for each waifu. Ideally going from some nice looking picture to a bust version and then to a 3D model. But, it's not very urgent. The other use case might be to get these models to be automatically animated, then we can make better videos of talking heads at some point. >Impressive The progress? Yes. The current quality? Hmm. Maybe if someone would build a more specialized model on top of it. The current results are still quite different from the input. >Share more If I make more, sure. You can make them yourself. It's easy. >Texture What for? Maybe it will smooth over some errors?
>>36295 >The progress? Yes. The current quality? Hmm. Maybe if someone would build a more specialized model on top of it. The current results are still quite different from the input. Heh, you put it like that then yeah I guess I meant the progress. I haven't seen a better single-image-to-3D-mesh-generation application thus far. BTW, can you export these meshes directly to a standard format from that tool? If so, then tweaking in Blender or Maya with smoothing tools should be quick & easy. Cheers, Anon. :^)
>>36296 Yes, it exports GLB files, which I used before. I don't remember the exact way, but they can be used in OpenScad and surely even more certainly in Blender.
>>36185 >This looks great but it's not clear to me if you can input many different pictures taken from different angles and then have it stitch those into a 3D object OR if it just uses one picture. >>36294 If you can run a model locally there are extensions/other models which allow for multi angled pictures to be used for generation. (Zero123++, InstantMesh Reconstruction, Unique3D, etc) https://github.com/MrForExample/ComfyUI-3D-Pack https://comfyuiweb.com/posts/best-3d-models-with-comfyui probably can find huggingface spaces of these models, idk haven't checked.
>>36304 Thanks, Axial! At some point hopefully before long, I'd like to begin modelling dear Sumomo-chan for design production. I have plenty of references for her, but if I could use AI to generate the basic mesh that would be a big help to kickstart the overall effort. I'm sure it will need re-topo'g (after all I want to animate it as a virtual daughteru first before the physical model), but still a nice headstart. >>36298 >Yes, it exports GLB files, which I used before. I don't remember the exact way, but they can be used in OpenScad and surely even more certainly in Blender. Great, thanks NoidoDev! That's really good to hear, and looking forward to seeing whatever you manage to do with this going forward. <---> Thanks for any tips and advice along the way, Anons! Cheers. :^) >=== -sp, minor edit -add'l resp
Edited last time by Chobitsu on 01/31/2025 (Fri) 07:54:48.
Open file (286.20 KB 675x623 Hunyuan3D_Test05.png)
Open file (246.08 KB 718x609 Hunyuan3D_Test06.png)
Open file (269.14 KB 649x609 Hunyuan3D_Test07.png)
I went on playing a bit with Hunyuan 3D >>36287, trying out the textures. It makes things look better. That said, I doubt that this is going to work for importing it into 3D modelling, since this is just a texture. For our main usecase of robowaifus only the White Meshes matter, imo. Still good to see how things progress, and it's good for content production and maybe also for some AI driven animated girlfriend app,
Open file (140.71 KB 718x609 Hunyuan3D_Test08.png)
Open file (165.60 KB 658x596 Hunyuan3D_Test09.png)
Open file (134.24 KB 658x596 Hunyuan3D_Test10.png)
Open file (153.40 KB 658x596 Hunyuan3D_Test11.png)
The next thing is to experiment with prompt to 3D. >stone bust of the head of a hairless young female, small cranium compared to the eyes, a sized down forehead compared to the eyes, the forehead should end not far above the eyes, very big eyes like Alita or a doll, smooth skin and lips, she is smiling, no eye lashes, no hair, no hair in the face, no hair around the eyes Btw, I can't upload more than one picture per drag and drop anymore.
Open file (136.11 KB 658x596 Hunyuan3D_Test12.png)
Open file (137.83 KB 658x596 Hunyuan3D_Test13.png)
Open file (123.20 KB 658x583 Hunyuan3D_Test14.png)
Open file (135.42 KB 658x583 Hunyuan3D_Test15.png)
>>36413 It's certainly necessary to work on this some more, but I consider this to be solved. What I really want though is picture to 3D, since then I can pic the face based on a picture. >stone bust of the head of a hairless young anime-like female, very big doll-like eye balls, smooth skin and lips, she is smiling, no eye lashes, no hair, no hair in the face, no hair around the eyes, the top of the bust cut of above the eye brows
Splendid research, NoidoDev. I hope I can sit down with this tool myself to get it running and do some of my own experiments with it for Sumomo-chan. Cheers. :^)
> (Janus-Pro [DeepSeek image generation] -related : >>36177, >>36179, >>36222 )
>>36435 This is great stuff. I love these pictures you churn out. Some of them anyways. The last set looks alien. AHHHHH! I wonder if you could use some open source software that meshes 2D pictures into 3D objects? Make 2D as you have, take pictures of the results and pop it into the other software and out comes 3D obj files. ????
>>36435 Thanks for prompts & examples. I've been messing with the huggingface demo. If anyone wants an easy way to install locally, this is on https://pinokio.computer/ , and they also have a low VRAM option (6GB). I got the image to 3D working locally but the text part had a compile error on my system.
>>36530 POTD Excellent resource, Barf. Have you been able to find any security issues with the tool's use yet? Cheers, Anon. :^)
>>36548 For pinokio, their install scripts are open like - https://github.com/pinokiofactory/Hunyuan3D-2 But, the app itself isn't hardened at all so all the local & temp files are unencrypted by default and all the servers aren't any more secure than your home network. CogStudio and AI Video composer on there are pretty nice too. EchoMimic is good for creating image-to-video avatars
>>36550 Thanks! Looks like I would need to use this version of the tool for my card: https://github.com/deepbeepmeep/Hunyuan3D-2GP >related: https://x.com/cocktailpeanut/status/1883198608477917622?mx=2 Cheers. :^)
>>36527 > The last set looks alien. AHHHHH! That's the ones with big eyes and more anime-like looks. I especially like the ones here on the right >>36412, if I try it again then to get something like this but without the hair. >I wonder if you could use some open source software that meshes 2D pictures into 3D objects? Make 2D as you have, take pictures of the results and pop it into the other software and out comes 3D obj files. ???? This is literally what this is all about, including other parts of this thread here >>32770 where I tried the same with exports from Meshy.ai. Except this here rather creates glb files for the meshes, I haven't even looked into the ones with textures. You can certainly change them into other formats.
Open file (281.16 KB 1084x709 3DBust.png)
>>36551 I just installed that and got the text to 3D working. Here's the first output using same prompt as Noido using the lowest ram option
>>36556 That's very good news. Though, obviously the prompt following seems to be more lacking (she has hair and small eyes). I would need to see more examples to know if that's consistent.

Report/Delete/Moderation Forms
Delete
Report