"Surprisingly, I am wearing clothes today; I usually don't wear anything at home."Yukishiro Mahiro
-
Lovable local shitposter and troll Jambalaya AKA God's Strongest Dragoon has won 1st place in the Phase Connect Clip Contest. Grats on the prize money and fishbucks!
Brave Group admits their data breach affects English applications too, expanding the number of potentially affected individuals into the five-digit range.
The forum project for a certain pink lagomorph mold golem has been completed.
The artist who created the public model Filian uses has complained publicly and to Good Smile that she does not have permission to make a Nendoroid of it.
AI Waifus and You 101
- Thread starter Watamate
- Start date
This is kinda giving me uncanny valley vibes ngl
I find it funny how Botan looks surprisingly natural in 3D while a lot of other talents are far more prone to the uncanny valley.
I didn't even know this thread existed.
re: @Tatsunoko Cool, but I'll be honest it just looks like a generic attractive female with a wig.
EDIT:
EDIT:
re: @Tatsunoko Cool, but I'll be honest it just looks like a generic attractive female with a wig.
EDIT:
Please don't trouble yourself, unless you really want to. Pippa in fancy clothes is just a funny image to me, no need to put a ton of effort into it.
EDIT:
Yeah, the pink dress is a much better match. Bit of a smirk in a couple of those too. Definitely an improvement.
Last edited:
Yeah, I don't disagree that the ears seem a bit pasted one. I admittedly rushed it since I had to leave for a few hours. Once I'm back I'll try to see if I can fine tune it a bit more.
Damn, I never got the alert for some reason, so I almost missed a couple of pages of very nice pictures. Was only your pippapost in the main thread that brought me back here.
Willing to share the models used? or is it a custom job for each character? So far I have not had very good results trying to get fully realistic looking scenes with the models I've used.
The model I'm using for most of my realistic gens is https://civitai.com/models/6424/chilloutmix with this vae https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main. And then I also throw in a LORA of the person I'm trying to recreate. Which you can usually find on civitai as well, or https://gitgud.io/gayshit/makesomefuckingporn#lora-list.
Seems I'm a bit out of the loop and have some catching up to do. Watched a couple of videos earlier and only just learned about civitai.com. Think I'll be spending a bit of time there for sure. Also in the process of installing LORA and will have a play with that soon enough.
Curious, anyone here used Elevenlabs or other voicecloning software? I've been playing the X-Series (by EGOSOFT, not the platformers by Capcom) a lot, and it has an incredibly verbose ship computer system that narrates everything like you're in an episode of Star Trek. It's also horribly generic and slowly turns into an audio sleep paralysis demon by the 1000th repetition. The X-series is known for being incredibly mod-friendly, so I was struck with the idea of voicecloning vtubers and using their A.I generated lines to replace the ship computer. Seems a fitting application of the technology.
It's what I used the Null April's fool voice clip for. It's a mixed bag, some voices it can replicate very well. others... not so much. This one used about 90~ minutes of him talking to train it. Along with turning down the stability quite a bit.
View attachment MATI.mp3
Following what Tatsunoko said, it definitely could work on replicating a voice either immediately or after some fine-tuning. I can't say much since so far I only used two samples (Portal 2 Announcer voice and Hina Sorasaki, acharacter from gacha game Blue Archive) and so far it's decent, so long that you only make them say words in english and more willing to toggle around with it.
Portal 2 Announcer saying railgun firing
View attachment UbOCtiBHNYr18cLGlY10.mp3
Blue Archive's Hina Sorasaki reciting the quote thingie from COD Black Ops 3 (also testing how a voice-clone that uses Japanese sample could be replicated to have it say english lines)
View attachment Hina ask you to listen to the sound of her voice.mp3
These aren't fine-tuned, but it's good enough as examples.
View attachment UbOCtiBHNYr18cLGlY10.mp3
Blue Archive's Hina Sorasaki reciting the quote thingie from COD Black Ops 3 (also testing how a voice-clone that uses Japanese sample could be replicated to have it say english lines)
View attachment Hina ask you to listen to the sound of her voice.mp3
These aren't fine-tuned, but it's good enough as examples.
One issue I can imagine with Elevenlabs is that it's now paywalled, needing you to pay at least $5/month to get the voice cloning and 30k letters to generate lines for it, with higher prices all the way to 'lets talk with us for the price' that has more and more letters to generate stuff and other benefits. That said, it's arguably the one that you can easily use if you don't have the willingness or time to dabble with python/other stuff to set up others voicecloning stuff such as Tortoise TTS or VALL-E, due to it being on a website.
Guess I might update this. So I kinda fucked up my PC while installing a few things. Will have to rewire the whole thing (in trying to fix 1 fan that was not working I ended up fucking up 3 fans + the CPU cooler). I have been meaning to get to it but my lazy ass hasn't really done much to fix the issue.
That said NOVELAI which was the program I was using for all of my AI got a huge update.
First Important Update:
1. You are no longer locked to 50 Steps when doing Image2Image :
The more steps the more the AI cleans the image. You can get good results with 20, and you can generate "free" images with anything up to 28 steps. If you used Image2Image they would lock you to 50 steps.
2. New Tools were added to Image2Image:
You could only do Image2Image before which basically tried to recreate any image + prompt and mash em together and depending on how much strength you added it meant that the image would focus more on the prompt instead of the image you imported. Low strength meant that the image that would come out would be almost identical to the one you uploaded. High strength meant that the AI worked its magic to change things.
3. NEW TOOL - Palette Swap:
Add an image, and change the colors but keep the image mostly intact.
4. NEW TOOL - Form Lock:
If you have an image with a pose you like you can use this to try to keep the pose as similar to the upload as possible and then you can change different things to it.
5. NEW TOOL - Scribbler:
For people with dogshit artistic skills you can draw a basic figure and the AI will work around that and turn it into something decent.
6. NEW TOOL - Building Control:
Seems to be used for designing architecture and other shit (I still haven't worked with it)
7. NEW TOOL - Landscaper:
Basically a tool to design backgrounds (I still havent worked with it either)
8. Generate more images:
You can make more images at the same time. Used to be a max of 4. Now you can do 1-8 at the same time.
9. More sizes:
You can basically make HD Wallpapers. It costs a ton of currency and you can only generate one at a time but the option is there.
Gonna post example of the scribble tool which honestly is a gamechanger if you are a terrible at drawing but you have an idea of something you want to see.
Drew this in MS paint in 15 seconds.
Prompt:
1girl, dark skin, dark-skinned female, long hair, large breasts, orange hair, blue eyes, crossed arms, navel, thong, skindentation, ahoge, parted bangs, {{nsfw}}, holding breasts,
Undesired Content (been using the same one I was using back in November):
(((deformed face))), (((deformed hands and limbs))), lowres, ((bad anatomy)), bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, ((blurry)), solo, (((censored))), mosaic censoring, pointless censoring, (bad proportions), (bad feet), (bad vulva), ((bad perspective)), futanari
As you can see it costs nothing to generate the image
>>>>>>>>>>>>>CURATED MODEL >>>>>>>>>>>>>>>>>>>>>>>>>> FULL MODEL >>>>>>>>>>>>>>>>>>>>>> FULL MODEL WITH NSFW TAG>>>>>>>>>>
Hands are still a bit fucky-wucky but honestly it gets the job done. When it comes to painting images and adding somewhat realistic shadows to clothing it is also a fucking godsend. You can paint a plain Tshirt on a character run that shit at a low strength with some fold lines and that shit will add the shadows for you.
Prompt:
1girl, dark skin, dark-skinned female, long hair, large breasts, orange hair, blue eyes, crossed arms, navel, thong, skindentation, ahoge, parted bangs, {{nsfw}}, holding breasts,
Undesired Content (been using the same one I was using back in November):
(((deformed face))), (((deformed hands and limbs))), lowres, ((bad anatomy)), bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, ((blurry)), solo, (((censored))), mosaic censoring, pointless censoring, (bad proportions), (bad feet), (bad vulva), ((bad perspective)), futanari
As you can see it costs nothing to generate the image
>>>>>>>>>>>>>CURATED MODEL >>>>>>>>>>>>>>>>>>>>>>>>>> FULL MODEL >>>>>>>>>>>>>>>>>>>>>> FULL MODEL WITH NSFW TAG>>>>>>>>>>
Hands are still a bit fucky-wucky but honestly it gets the job done. When it comes to painting images and adding somewhat realistic shadows to clothing it is also a fucking godsend. You can paint a plain Tshirt on a character run that shit at a low strength with some fold lines and that shit will add the shadows for you.
Honestly for making thumbnails this shit has been a godsend (I used to manually edit fanart to make them and it was a huge pain in the ass plus I always run the risk of some moral busybody looking to cancel me for the crime of "EDITING art".
Case in point these thumbnails were made all made by editing already existing art
Not really vtuber related, but I saw this clip someone made, and as far as AI-Generated (IMG2IMG) videos, it's probably one of the cleanest I've seen. Some frames definitely still have fucked up hands and it's still inconsistent, but it's crazy to see how far we've come in a few years.
View attachment ez39xa.mp4
View attachment ez39xa.mp4
Last edited:
Oh yeah I saw that as well. Already it's a lot less janky than the last animation I saw (thought admitedly I've not seen much). It's definitely going to be interesting times ahead.
Also, I can't believe II never heard of civitai sooner. Maybe it was a curse because I've spent most of my free time the past few days downloading, cataloguing and using over 100 checkpoints and LORAs. Mostly non-tuber related, but here's a bit of hag appreciation I whippped up just now showing the same prompt/seed with a few different models:
Last edited:
Don't worry, Japan has you covered.
It's got Gura in a bikini, so I'm spoilering it. IMG2IMG running over the top of an MMD animation.
Where I saw this - reddit
Is this something that vtubers could use as a cheap 3D? There is mocopi but you still need a 3D model. Just curious as I know jack about either.
Similar threads
