Artuis is a versatile and flexible model, it allows you to generate crisp, high-resolution images:
This is a demo of NSFW capabilities of my model, you can check main model page in my models list.
Keep in mind that this is an SD 2.1 768 model and its NSFW capabilities are limited compared to SD 1.5 models.
To make unique faces, mix celebrity faces that the model knows: [emma stone : scarlett johansson : 0.5]
If you liked my model, please leave a review, it helps other users to find this model and see its features!
This is SD 2.1 based model, also I have SD 1.5 based version with more NSFW capabilities: https://civarchive.com/models/47691/artiusv1-5
The model works best with the following samplers:
Euler a
DPM2
DPM++ 2M Karras
Optimal settings:
Clip skip 1
Eta noise seed delta: 31337
DPM++ 2M Karras
25 Steps
CFG Scale 6
Resolution from 768 (768, 1024, 1280)
The model is prompt sensitive, so it depends on your imagination.
I've generated some demo images, you can study the prompt and make your own based on them.
Description
N_S_F_W support
Intergrated VAE
Improved anatomy
Improved sharpness
Improved eyes
FAQ
Comments (22)
Gentlemen, The maker don't like to give girls a real cat. LOL~!
I just came here to see if this was a non NSFW model and I had just forgot or something just to see all the boobs which confused me since I generated an image with accidental nudity but the woman was like a barbie doll below her waist and it was so weird lol
SD 2.1 Bro
It's very difficult to get this model to generate a proper vagina. At first I thought it was adding some kind of icon to censor the vagina, but then I realized it was just very confused about how to render one.
Am I the only one withe the error?
OutOfMemoryError: CUDA out of memory. Tried to allocate 9.00 GiB (GPU 0; 12.00 GiB total capacity; 2.07 GiB already allocated; 7.60 GiB free; 2.17 GiB reserved in total by PyTorch) If reserved memory is allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Time taken: 26.84s
Torch active/reserved: 4286/7446 MiB, Sys VRAM: 11992/12288 MiB (97.59%)
How running out of VRAM is related to this model ? lower your settings or get xformers.
@madPsyko I have no idea... works fine with other models, which settings should I lower? I do have xformers, here is the log from startup script:
`Launching Web UI with arguments: --xformers --medvram`
@aleksml_84 hi, try this:
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
@mech4nimal I tried it and had a different error, googled the error and it looks like I used some v1.5 embeddings which are not compatible with v2, thanks.
it is said, 2.1 model needs a .yaml file，the author may forgot it
When I tested the same prompt I got the identical image both with this and the non-NSFW version of artiusv21. I'm not sure if used or did not use a VAE in either case (think not, none in file info mentioned). the only difference is that thenon-nfsw version output is sharper/more detailed.
Thanks for the Artius Models. They have great potential imo. What could be better: Hands, Eyes (seem often to have a kind of drawing style around the iris) and last - but not least - nipples ;D
I don't get it... How is this a legit SD2.1 model? I thought the 2.1 stuff was 'off limits' as far as merging in NSFW? If this is possible why aren't there a dozen 2.1 models like this?
Because this model is additionally tuned on the NSFW dataset
Hi Sanchez,
Are you working on any updates to this model? This is by far the best one I've used until now, but there is one thing I dislike about it: it tends to make the faces too masculine, with a large forehead and pronounced cheekbones, straight brows etc. Other than that it is magnificent. Maybe you could point me on how to use negative prompts, because it seems to ignore (large forehead/big forehead), (masculine) etc. please.
Thank you,
Tibi
ive seen that on many sd2.1 models hence why so few people use them. it seem they tend to make trans more. so you have to heavily throw tags into negative prompts to combat it. sd 1.5 is still way better with more freedom and more extras made by the community
Have you experimented with inpainting using another model?
@vegetarian_crocodile Yes, I have. Sometimes the results are credible, but most of the times you could tell there was a region "inpainted" over the original. I've managed to tame this model with some negative prompts, but more tinkering is needed.
think you meant to tag this as "NSFW" "Not Safe For Work"
<thumbs up>
lol, its tagged nswf, not safe work for
A model that can generate only Females naked IS safe for work. Thats not real nsfw. come back to me when you have dicks then it would truly be nsfw.
this softcore model of poses lol. boobs arent pornography.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.