PLEASE LOG IN OR REGISTER TO BE ABLE TO SEE THE PICTURES AND ATTACHED FILES!
|
Author |
Message |
suomynonA
Warnings:
Posts: 0
|
|
Quote |
Sadly, Waifu Diffusion 1.4 isn't the AnythingV3 killer it was touting itself to be. I've read that their enhanced training will be significantly improved in 1.5, but I think I'll err on the side of caution.
That being said, it's not terrible. Like WD 1.3, it has more of a flair of creativity, but relies on negative prompts to smooth out the... Ah... Anatomy. As for the naughties, it seems to be a bit better at lower anatomy in some cases. You can also get characters to NOT look at the viewer, something Anything V3 struggles with.
There was a false AnythingV4 released on Huggingface. Just a merge blend of AnythingV3 taking the name. There are quite a few blends out there now, as well as specialized models on various tropes of hentai. No specialized loli model that I know of.
|
|
Tue Jan 17, 2023 12:15 pm |
Profile PM
|
|
davidboy
Hikki Angel
Warnings:
Posts: 6440
|
|
Fri Jan 20, 2023 6:24 am |
Profile PM
|
|
davidboy
Hikki Angel
Warnings:
Posts: 6440
|
|
Quote |
Some with forest background, and some others
|
|
Sat Jan 21, 2023 11:51 pm |
Profile PM
|
|
davidboy
Hikki Angel
Warnings:
Posts: 6440
|
|
Quote |
more
|
|
Sun Jan 22, 2023 12:00 am |
Profile PM
|
|
Lolibop
Warnings:
Posts: 0
|
|
Quote |
I hope this becomes easier to install over time. I would very much like to make my own catgirls via AI
|
|
Sun Jan 22, 2023 4:17 am |
Profile PM
|
|
suomynonA
Warnings:
Posts: 0
|
|
Quote |
Lolibop: I hope this becomes easier to install over time. I would very much like to make my own catgirls via AI
hxxps://www.reddit.com/r/StableDiffusion/comments/zpansd/automatic1111s_stable_diffusion_webui_easy/
Then for models, most are available from hxxps://civitai.com/ , AnythingV3 is by far the most user friendly and highest quality (In my experience thus far). Just use Danbooru style tags separated by commas in the order of importance. If given the option, always choose a safetensor file over a ckpt. Same goes for f16 vs. f32, though some video cards aren't able to use f16. Also, favor 'pruned' models if you have no interest in doing trainings, essentially it cuts out that part of the model but has no effect on your generated images.
Some models (Notably AnythingV3) have a supplementary 'Vae' file that goes along with it. Essentially it's like a mini helper model that cleans things up a bit.
Also, I did try the fake Anything V4 and V4.5. They're not terrible, but they lose a lot of the 'drawn' look of AnythingV3. Keeping it simple, prompt was: 1girl, loli, nude
|
|
Sun Jan 22, 2023 1:19 pm |
Profile PM
|
|
suomynonA
Warnings:
Posts: 0
|
|
Quote |
Okay, I got bored/inspired and did some Img2Img.
See if you can guess where the base images were from. (They're all screen caps of games.)
AI still isn't great at fine details from far away... I did my best.
Edit: hxxps://www.reddit.com/r/StableDiffusion/comments/10kwln8/automatic1111_webui_binary_v100pre_released_no/ Here is another possible easy path into using Stable Diffusion.
|
|
Tue Jan 24, 2023 12:55 pm |
Profile PM
|
|
davidboy
Hikki Angel
Warnings:
Posts: 6440
|
|
Quote |
some
|
|
Thu Jan 26, 2023 9:20 am |
Profile PM
|
|
SConstance
Warnings:
Posts: 0
|
Some of my creations |
Quote |
I have a lot more but this is just a small selection of what I've made
|
|
Thu Jan 26, 2023 5:55 pm |
Profile PM
|
|
OG_SOLF
Warnings:
Posts: 0
|
Re: Some of my creations |
Quote |
[quote="SConstance"]I have a lot more but this is just a small selection of what I've made[/quote]
Great Art!
|
|
Sat Jan 28, 2023 1:05 am |
Profile PM
|
|
davidboy
Hikki Angel
Warnings:
Posts: 6440
|
|
Quote |
some
|
|
Sun Jan 29, 2023 8:52 pm |
Profile PM
|
|
Lolibop
Warnings:
Posts: 0
|
|
Quote |
Ok so... I have it installed now, but I'm having a little trouble with the commands needed to install the Civtai API. The github is very vague.
suomynonA:
Lolibop: I hope this becomes easier to install over time. I would very much like to make my own catgirls via AI
hxxps://www.reddit.com/r/StableDiffusion/comments/zpansd/automatic1111s_stable_diffusion_webui_easy/
Then for models, most are available from hxxps://civitai.com/ , AnythingV3 is by far the most user friendly and highest quality (In my experience thus far). Just use Danbooru style tags separated by commas in the order of importance. If given the option, always choose a safetensor file over a ckpt. Same goes for f16 vs. f32, though some video cards aren't able to use f16. Also, favor 'pruned' models if you have no interest in doing trainings, essentially it cuts out that part of the model but has no effect on your generated images.
Some models (Notably AnythingV3) have a supplementary 'Vae' file that goes along with it. Essentially it's like a mini helper model that cleans things up a bit.
Also, I did try the fake Anything V4 and V4.5. They're not terrible, but they lose a lot of the 'drawn' look of AnythingV3. Keeping it simple, prompt was: 1girl, loli, nude
|
|
Mon Jan 30, 2023 4:35 pm |
Profile PM
|
|
suomynonA
Warnings:
Posts: 0
|
|
Quote |
You shouldn't need an API for civitai. Civitai is where you download the models you want to use. In short, a model is kind of the 'genre' of AI generation you want to use. There are ones for anime, more artistic ones, furry, etc. You download the .ckpt file (Ideally .safetensors) put it in \stable-diffusion-webui\models\Stable-diffusion then you can use it in WebUI.
I haven't used the installers at all, so if that's a part of it I'm unfamiliar with it.
Last edited by suomynonA on Tue Jan 31, 2023 1:03 pm; edited 1 time in total |
|
Mon Jan 30, 2023 9:05 pm |
Profile PM
|
|
NoxL0rd
Warnings:
Posts: 0
|
|
Quote |
suomynonA: You shouldn't need an API for civitai. Civitai is where you download the models you want to use. In short, a model is kind of the 'genre' of AI generation you want to use. There are ones for anime, more artistic ones, furry, etc. You download the .ckpt file (Ideally .safetensors) put it in \stable-diffusion-webui\models\Stable-diffusion then you can use it in WebUI.
I haven't used the installers at all, so if that's a part of it I'm unfamiliar with it.
Incredible stuff! What settings and prompts do you use? I use anythingv3 but cant generate anything to this quality
|
|
Mon Jan 30, 2023 11:04 pm |
Profile PM
|
|
suomynonA
Warnings:
Posts: 0
|
|
Quote |
I actually do very little. I keep the prompts pretty simple using just Danbooru style tagging in the order of importance. Using The 'PNG info' tab in WebUi, you can actually get all of the generation information on anything I've posted thus far. Really goes to show just how far ahead of the curve NovelAI was/is that models based off of it are so powerful without really needing much or any negative prompting.
DPM++ 2s a Karras is good to speed things up since it can do very good quality at 5-10 steps. There's actually a new thing that supposedly can make 25 step pictures in half a second on Nvidia cards.
These days there are models that do the more naughty stuff a lot better than AnythingV3. There is an extension you can easily get that add in an 'auto-complete' style thing when putting in your prompts that helps quite a bit, especially if it's something that uses parenthesis (I.E. Metal_gear_solid_(series)) since it will properly format it to utilize it rather than making it an emphasis.
|
|
Tue Jan 31, 2023 12:33 am |
Profile PM
|
|
|
|
You cannot post new threads in this forum You cannot reply to threads in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum You cannot attach files in this forum You cannot download files in this forum
|
Do not upload or mention any illegal content. Violators will be reported to the authorities. You are responsible for your uploads and actions.
If it is illegal for you to view adult lolikon material in your current location, leave this site immediately.
LAH is in compliance with DMCA.
|