Make sure to use trigger words, "leg stumps" & "arm stumps" as appropriate. In supported versions, "shoulder stumps" and "thigh stumps" can be used for shorter stumps cut off at the shoulder/thigh. "amputee" and "quadruple amputee" were NOT used often in captioning training images, so they may not be very effective.
Recommend using 0.8a if using civitai generator. For later versions, try using locally, or try using a merged model like autismmix. dora versions after 0.8a are mostly just fiddling with certain details: in particular, trying to get individual arm stumps and leg stumps to generate independently and consistently for non-quadruple-amputees, improving stump cover generation, improving chain details, etc. If you just want quadruple amputees, use 0.8a.
Prompting guide:
stumps:
For most versions, "arm stumps and leg stumps" will work best, as the training set has the most examples of quadruple amputees. Other stump combos work fairly unreliably, and may need additional strength (e.g. (arm stumps)).
stump options:
arm stumps and leg stumps
quadruple amputee
works the most reliably
arm/shoulder stumps
amputated arms or shoulders
leg/thigh stumps
amputated legs or thighs
right/left arm stump
amputated arm
right/left leg stump
amputated leg
stump covers (non-exhaustive list):
Stump covers can be optionally specified. Some work better than others, depending on the version.
stump cover options:
bandaged
[chained loops/handles/eye bolts on] flat/rounded [red/orange/yellow/etc.] [metal/cloth] caps on
scarred
eye bolts in
For prompting, combine stump cover options with stump options. E.g. "bandaged arm stumps and leg stumps".
Largely trained on nsfw images, so include specific clothing items in your prompt if you want clothed images. Clothing covering/obscuring stumps will not work as well as sleeveless short-sleeved clothing. Working on expanding dataset for greater flexibility of which body parts are amputated.
Description
FAQ
Comments (31)
Just found this and started using it for some img to img work. So far, so good. It won't work with anything but a Pony checkpoint, which is a bit of a drag. Still, it mostly does what it is supposed to do. The only real issue I have with the Dora is that it seems to default to nudity. I have been using it mostly about 1.4 strength to ensure I get quads and I can't get a single person to wear clothing but on rare exceptions. Frankly, it's the best way to get amputees I have found. Good job, Out_of_username_ideas. Maybe future iterations could have a few more options and be usable with SDXL. That would make this already high quality Dora even better.
EDIT: All the newer versions have improved this DoRa. I am getting clothing and other options whenever I use it now. It's very good.
Thanks for the feedback! The dora was originally intended for nsfw generations primarily, so most of my dataset is nsfw. Since it's also all drawn images, I trained on the pony checkpoint.
I'm working my way through captioning a bunch more images I've collected, some of which are not quite so nude, so stay tuned! Not sure when the next version will come out however. I have a bunch of images to tag. I will say my expanded dataset still is largely lacking clothing with sleeves for arm amputees however. I was worried that would confuse the concept.
@out_of_username_ideas Yeah, I would not be surprised at it confusing the AI. I've managed a few images that are pretty good using your Dora. Right now I am messing with Img2Img and started out the amputee thing using the Human Onahole and then Img2Img to make it more realistic. I will tell you I am particularly fond of your Dora because it actually leaves the stump. It's a bear to get a sitting amputee that doesn't look like it should fall right over. If it would help, and my images are not greatly refined as I am very new to AI generation, I can send you some images for your data set. In trying to prefect things I have made a lot of images - some worth publishing and some not. And, yeah, some with clothing!
@hdean Thanks for the offer! I'll have to decline unfortunately. I like to keep ai generated images outside of my dataset to avoid potentially introducing and ai-specific artifacting or something like that.
On the stump thing, yeah, that was part of my reason for making this dora. A lot of the loras I searched for seemed to just leave a completely smooth surface behind, at least in the sample images.
@out_of_username_ideas An update already? Nice. Gotta tell ya, I really like the Dora. I've done quite a bit of fiddling with it. While using an SDXL Checkpoint can muck up the image, if I use SatPony for the initial image and then run it through an SDXL checkpoint with the Dora in a Lora loader at low strength it keeps the SDXL checkpoint from adding limbs. Also,not used the Dora as a stand alone, yet. It's been img2img exclusively. So, I don't really have a great data set. I do use it at a high setting, though. generally between 1.5 and 2. It's freaking great. I will get a couple images up for your gallery when I get the chance.
Tell me, does the newly updated model have clothes? I've a plan to use it in conjunction with another lora to make teddy bears.
@hdean Not sure if the new version does any better with clothes tbh. There are a few more clothed images in the dataset, but not a whole lot. I mainly trained it as a test to see if expanding the dataset would improve the dora quality in general. So far, there are some minor improvements in less overtraining for certain caption terms, though it looks like there's some style contamination for males. I still have a bunch more images to caption however.
I will say that as long as you include some items of clothing in your caption, they should generally show up? Default is still largely nude though.
For teddy bears, I have no idea how that'll go, but sounds like a fun idea! I feel like you're more likely to get some body horror then cute teddies however.
@out_of_username_ideas Believe it or not, I have managed two good images so far. Still need to get it right. I think img2img with the hair lora I am messing with might work, so long as I can start with a fully clothed limbless woman. Wish me luck.
@hdean that's great! hope it works out for you
@out_of_username_ideas I am already ahead of the game thanks to your Dora!
Holy crap! I hadn't used the new version until just now! V6 is freaking powerful! I had been using the original at about 1.5. When I loaded the new version it turned my image into a cartoon. Dropped the level to 1 and got the exact sort of image I was looking for - fur and all. Got an amazing image! Whatever you did really did a number on the original Dora! Great freaking work. I am gonna run a few more and then (later) plan to load up an image for your Dora page.
@hdean That's great! I actually had mixed results when testing the new version tbh. I was debating renaming it v0.5b instead lol. I actually forgot to flip augment a bunch of my training images too, so I might release a new version ahead of schedule, if there's an improvement to be had there.
@out_of_username_ideas FYI - for my original img2img I am using dpm++sde_gpu with karras and then filtering the image through an SDXL model for the final run through. The images I am getting are pretty damned good. Mostly not what I want but that's mostly because I am struggling with the fur lora. Also, your Dora tends to give some very cut, muscular images. Will be trying your Dora as a stand a long a little later today. But the img2img are amputees based on your Dora. So far as I can see, the only thing that might make this a better Dora is if it can be run through an SDXL checkpoint. But I have no idea even how to do what you're doing. I am fairly new to AI generation and have barely scratched the surface.
Ironically, I recently got a new computer. However, I had not gotten into AI and just needed an upgrade. Since I don't play video games or anything, I never bothered to get a top of the line GPU. Now, I wish I had, 'cause I have just about lost my mind playing with Comfyui.
FYI - I don't know how I found your Dora, but when I look for it now on CivitAI I cannot find it. Something is out of sorts.
@hdean I know the feeling. I decided to play around with stable diffusion on a whim a year or so ago and got hooked. I'm less obsessive now on generating images, but creating models is still satisfying every now and again.
On searching for my model, try checking if you have any filters enabled on the left. If that doesn't work, try checking the content controls/content moderation settings on your account settings.
@out_of_username_ideas That did it. It pops up now! Still, I wonder how I found it in the first place.
You are one busy creator!
@hdean lol. Just trying out the low-hanging fruit of flip augmentation. Next update probably won't be for a while unless I stumble across a good process improvement or get suddenly super motivated to tag a whole bunch of images.
Quick thought for you. I've noticed that all the amputee LoRas seem to have issues with any position not sitting upright or lying on the back. I've been fighting to get any of them to show an amputee with stumps crawling or on their side. I know I may be sounding pushy, but I am hoping you might consider an updated version with just such models.
@hdean Whoops, completely missed that comment.
I'm afraid for the "on the side" position, I've got maximum 1-2 images of that in my dataset. It's not a position that seemed to come up in my searches when collecting, so dunno if that'll ever get working well.
For "crawling", I've got a similarly small dataset. I've got a couple of "on all fours" images however, so that part at least should work.
@out_of_username_ideas Hope the holidays have treated you well, and you have a good new year on the horizon. I see you've a newer version of your DoRa, which I plan on testing soon. So far, I have loves both of the models of yours I have used - the Amputee and the "In Wall" DoRa. Gonna give your newer version a shot very soon.
I had a thought the other night, as regards your DoRa, and thought I would throw it your way. Of the things I see lacking in most of the amputee models is a consistent way to end the stumps. I've managed to get a sort of plate at the end of the stumps, but it might be a groovy thing to have spherical ball joints to end the stumps.
By the way, since you are doing all these DoRas, I assume you've a place you post your own art. I would be interested in seeing what you produce.
@hdean Thanks! Hope you've had a good holiday season as well. I'm glad you've enjoyed my doras. On stump caps, it's been a while since I've worked on this dora, but I believe I don't have any examples of ball joints in my training data unfortunately, tagged or untagged. Beyond the caps shown in the gallery, I think I also have a few examples of sockets on stumps for prosthetics and the like, but not sure how well it works since I had only a few examples of those. If you've got any recommendations on where to find any, I'd love to know.
I've been working on other doras recently, so not sure when I'll get back around to this one. I do plan on reworking this dora and my stuck in wall dora sometime in the future however, since I've learned a bit more about training doras in the meantime.
For my own personal art, civitai's the only place I post as of now. Ironically, I spend much more time working on training doras now than actually making images with them.
EDIT: took a quick look at my training data and it looks like I do have a few images for "eye bolts in arm stumps and leg stumps". It doesn't generate too reliably, and is usually paired with chains however.
@out_of_username_ideas I googled dolls and ball joints and found a bunch of confusing things. Basically, think of a tow hitch. The ball would fit into an accessory - in dolls, to the arm. I had assumed a lot of your DoRa samples were hand drawn. Am I incorrect?
@hdean Got a picture example? Right now, I'm thinking of a rod with a sphere on top of it coming out of the stumps.
For my training set, most of it is digital art, with a few 3d renders.
@out_of_username_ideas I can't find a good image, but your description is right on the mark. Just a little post extending with a ball on the end. I think it would make a good addition for the DoRa.
Sorry about the 5 day wait: I've been pretty busy.
@hdean reviewed my dataset and unfortunately, I couldn't find any examples of ball joint stumps. If you ever come across one, let me know, and I'll give it a try.
@out_of_username_ideas Will do!
Did your amputee Dora get caught up in the great Civit purge?
@hdean wdym? Isn't this comment on this dora page?
@out_of_username_ideas Actually, it wasn't. I did a search for the Dora and couldn't find it anywhere. I was able to comment to you based on our conversation notifications. Oddly, I just checked one of our older messages and was able to find it. But it didn't come up in search. Could have been a glitch.
@hdean You might have a filter on. I'm able to see it in search.
I think it was a temporary error on the site. I checked my filters. I can see it now. Just that one day it wasn't available.