steps: 30, cfgscale: 5, sampler: dpmpp_3m_sde_gpu, scheduler: karras/exponentional, vae: sdxl_vae,
positive: (score_9), score_8_up, score_7_up)
negative: score_6, score_5, score_4, score_1, source_anime, source_pony, (worst quality:1.2), (low quality:1.2), (normal quality:1.2), lowres, bad anatomy, bad hands, signature, watermarks, ugly, imperfect eyes, skewed eyes, unnatural face, unnatural body, error, extra limb, missing limbs, painting by bad-artist, cross-eyed, (lazy eye), bucktooth, wrong number of fingers on hands,
Description
FAQ
Comments (23)
sry for low resolution pictures, i copied them from SwarmUI, not directly :D
What are all those wild 3 letter combinations in the negative?
@LetTheBassDrop i dont know i found it in one pony model on this site, there is comment where is it explained
@LetTheBassDrop there are translation lists online to filter. The 3 letters are basically trigger/key words for a certain artist style.
Model is actually good.
The ups
GREAT eyes, even without ADetailer the eyes are superior to Vivid's
Good hands, around 60-80% of hands generated are spot on, plus the nails somewhat match both hands.
Prompt understanding, Vivid's has trouble with backgrounds, this one does not, it generates almost entirely different backgrounds everytime PLUS the prompts crowd and crowded seen to work 80% of the time.
Great facial expressions, so far tested two, blowjob and neutral, neutral actually looks 'realistic' and blowjob the vaccum shape is insanely good.
The downs (depends how you look at it)
To much blur at the BG's, great for POV's but standing poses can be a problem.
Quite 'grainy', it has that Bing aesthetic to it, some like it some don't.
Starts to fail at far poses like not understanding stabelished characters, see Cynthia for example.
Besides that it is REALY good.
i agree with everything.. thx for your time and test
Just to add, this model is INSANELY good at prompt understanding, I already can see I can pull some goofy shit with it lol, props to you, great model my favorite as of now lol.
That good?
Works extremely good with my gaping pussy LoRA. it translates the LoRA exceptionally well and adds great variations of poses to it. Love it so far!
Model looks awesome, thanks for sharing. Baked in VAE would be phenomenal for fooocus... any plans on doing that?
i have very slow pc, when i try bake VAE in a1111 my PC froze....
@vendo19324 Ah, I see. I have the same problem.
@vendo19324 if you have plenty space on you SSD try Increase your Virtual memory or File page system to 64 GB, that maybe help you with the frozens, Nice Merge dude
@vendo19324 which VAE do you want baked in?
I can bake it in, but keep in mind its done with rocm, no idea if that changes anything. (I already baked myself one into your modelf or myself, but I think it changed seeds.)
Wow, this is amazing! Thank you!
I am getting a lot of females even with male only prompts. Looks tranny like with earrings. How can I avoid it?
@CyclopsGER i did over 100 images and i dont have this type of problems, can you post your prompts?
@vendo19324
Pos:
photo, raw,realistic, mature elderly grandpa chubby bald homeless polish (man:1.3) sitting leaned back with wide spreaded legs and showing his anus, large penis, legs up, chubby thighs, looking at viewer
Neg:
(female:1.2),glasses,score_1,low-res,bad hands, deformed, extra fingers, low quality, moire, overexposure, poor quality, jpeg artifacts, censored, low resolution,3d,(more than 5 fingers)（more than 6 toe),(4toe),render,cartoon,broken,bad anatomy,censored
@vendo19324 Not every image is female but a lot of them. Generated thousands similar.
@CyclopsGER maybe try change grandpa, i guess model is not familiar with this word very well, try older man etc try synonymus to this word
Is there a way to prevent trained bodies or is this a Pony problem? I try to make ordinary people without a lot of muscles.
Easily the best model I have ever tried
How can Lora learn from this model? I'm trying to learn but getting errors