style is more anime like on some concepts and is overall more bright
What is better:
- much much more stable
- slightly better artist tag understanding is back (nonLora 1.1 remains better at it)
- it is a strait up update of mostly everything (i don't lie)
What is worse:
- Backgrounds changed much, feels more illogical, but is much better at certain backgrounds in compare to pony
- Can be to bright for some
2024-05-15 03:48:07 +0000 UTC View PostUse this as default model.
Whats better:
More anime like with much less realistic looking faces
Much stronger stable model art style
Less detached limbs
Some characters working better then before
Higher output quality (more like hassaku SD1.5)
Remains the same:
Hands remained mostly the same
SFW remained mostly the same
Note: censored in nega...
Use this model if you want to use artist tags or for lora merges
It is not as stable then the lora merge and produce different art styles on use
Note: censored in negative tends to output more realistic looking images. But most created images are in default not censored
2024-04-15 21:36:17 +0000 UTC View PostMid state progress model.
It is a bit unstable in overall style, some concept having realistic styles and on others 2d or a mix of it. I see that training normalize it overall, probably ponys sources that goes wild a bit.
It is more easy to handle in compare to pony and much more anime like, with all strenghts of concept understanding of pony.
What i see on testing, that i want to correct:
- Sometimes face art style does not match body art style.
- Does have sometim...
Trained with 1,2m images over 1 epoch.
But because its only one epoch + extra small eastetic runs with extra concepts, its better at mostly everything but only slightly. Most needs more training. sfw is a bit more stable.
2024-02-22 20:10:55 +0000 UTC View PostTrained on 250k explicit images
With a low LR for only one epoch.
It makes nsfw a bit more stable, without changing to much.
Planning to do training on much more images, including sfw images for longer, but my datasets are not ready yet.
2024-01-27 08:14:24 +0000 UTC View PostMy current SDXL training state. Not a merge like 0.1 before.
That means, its not much stable in style and colors in favor of nsfw concepts.
Seen also some other issues with it that i want to correct in further versions
What is better:
- Better constructed backgrounds and clothes
- Hands are slitly better
What is better/worse for some:
- a little bit more realistic, mostly noticeable on the eyes (a bit smaller in compare) with more visible nostrils.
Tested on clip2 with mostly CFG Scale 7 to 8
2023-06-29 22:34:42 +0000 UTC View PostSmall update on those things:
- Slightly bigger breasts in compare on prompts large and huge breasts on some images
- Slightly better formed hands midrange (from bad hands to slightly better bad hands... on some images it helps)
- Better response on prompt fellatio with a higher probability, that the tip is in the mouth
- Details on clothes or bodys (like wings) are less likely to be detached from bodys and is more equal right and left
- Le...
2023-06-22 17:14:05 +0000 UTC View PostWhat is better:
Style is mostly the same to hassaku, but not with a nsfw focus, but a focus on stability and image quality
2023-05-30 22:54:32 +0000 UTC View PostWhat is better:
- Slightly more open pussy and anus (with highress fix)
- Anuses are less forgotten
- More responsive to nsfw tags
- Less realistic faces, without face defining prompts (ugly big lips faces, not good semirealistic)
- Slightly less changes of double limps
What is better/worse for some:
- A little bit less "shiny", more drawn and less extreme bright skin.
- Clothes are more revealing
What is worse:
- More chances of dou...
2023-04-26 22:26:37 +0000 UTC View Post
Please look at the channel "what-i-do" to inform you, on what i currently working. Patreon is not the best for daily update feeds.
It also where I post announcements, updates and polls!
2023-04-18 08:26:50 +0000 UTC View Post
Added more nsfw into the model layers.
What is worse:
- Eyes have on some occasions more problems to form pupils (if at all).
- Motion lines are more likely created and blend on highress fix into the outer lines and create "stepp artifacts and sparkly skin effects". Negativ "Motion lines" helps sometimes.
What is better/worse:
- On nsfw prompts, arms and hands are more likely to go down to the insertion for masturbation or spread.
What is better:
- pr...
2023-02-22 22:16:29 +0000 UTC View PostModel created to be stable as possible. Less different eyes, less extra limps. Cohesive backgrounds.
It has the same lewdness like abyssOrange2 hard. But it doesn’t have it in it. Because of that, it does not have the lighting of it and is less realistic.
If you know bloodOrange and elyOrange. It is something between. But mixed in a way to get a stable result with nai - naisfw and gape60-nai
AnythingV3 + ElysiumV2 + (NAI - NAISFW) + (gape60-NAI)
No vae included.
<... 2023-02-09 00:18:05 +0000 UTC View Post
A merge from scratch, with the methods of AOM2.
Model mixed with the maximal lewdness in mind. Model is so lewd, it has tendencies to transform background stuff simply to sex related stuff.
No vae included.
Details:
AnyGape + basilMix
AnyGape unNetMerge with basil: 1,0.9,0.7,0.5,0.3,0.1,0.5,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0,0,0,0.05,0.15,0.25,0.35,0.45,0.5
0.2 add differences: nai - naiSFW
0.1 add differences: anyGape - anyV3
2023-02-07 08:34:58 +0000 UTC View PostSet up a discord server, i‘m new to this. So help and tips would be great!
2023-02-07 01:56:19 +0000 UTC View PostPlanning to create a discord server, so we can appreciate your images you create.
Images of all models are allowed. Sharing loras and talk about it.
2023-02-06 13:20:55 +0000 UTC View PostIn short: with enough support, i can faster update my model.
Now my humble pc is the only method, to update my model. It took a long time, mostly to check varius generated images on it.
I already use google colab to fasten that process. But I need more calculations units on it and they are expensive. But those google V100- or A100-Nvidia-GPUs are really sweet.
It is a dream for now, but I want to transition from simply merge my model to training my...
2023-02-06 12:51:19 +0000 UTC View Post