Update notes here :
H-2A2b - Consistent 1/2 loli-generating model, Images uses quality prompt mostly no loras.
Blessings of Yig Yig-B5CIP2 - A more likey-stable, consistent, effective merged model.
Blessings of Yig A1A - You are "blessed" with an inconsistent model that i merged randomly.
C2Alpha-DpB-B2 - Slightly changes, re-merge with IL2A4 and T6AO4.5 for similar style and details.
T6AO4.5 - A newer redo of T6A4 Raw, re-merge with IL3C2 and T6A4 for better detail, of course the rest is prompt-depended.
Illustrious, Different style, LoRA-merged in, it works with or without negative prompts ( maybe ).
Settings from my Images :
~ Euler A, 8-12 Steps, 2.5-3 cfg
~ No Hires.
~
~ DPM++ SDE works : 5 steps , 3 cfg ( Be wary of the Results )
~ Sometimes LoRA-merged in. But not all the time.
For some quality prompt of my suggestion :
< Prompt Insert Here > ,
BREAK
nsfw,
ambient occlusion, raytracing, best lighting, blum effect, vibrant colors, depth of field, absolutely eye-catching, intricate cinematic background, (beautiful face),
BREAK
masterpiece, best quality, highly detailed, ultra-detailed ,beautiful eyes, big eyes, detailed eyes, detailed pupils, perfect body, dynamic pose, dynamic shot, detailed illustration, very aesthetic, sharp focus, dynamic shot, dynamic pose masterpiece, best quality, ultra-detailed, very aesthetic, sharp focus, depth of field, vibrant colors, ray tracing, best lighting, detailed illustration, detailed background, cinematic, beautiful face, beautiful eyes, blum effect, highly detailed, ambient occlusion, intricate cinematic background, best lighting, < Prompt insert here >< Prompt insert here >,
BREAK
masterpiece, best quality, ultra-detailed, very aesthetic, sharp focus, depth of field, vibrant colors, ray tracing, best lighting, detailed illustration, detailed background, cinematic, beautiful face, beautiful eyes, blum effect, highly detailed, ambient occlusion, intricate cinematic background, best lighting, sharp focus,Description
A newer version of T6A4 Raw...
FAQ
Comments (13)
The most perfect model in my mind!
Thank you !!
This model is a sleeper hit, great generation results, very versatile
el mejor modelo que encontre
AMAZING!! Hands down instant favorite - handles long prompts well and the images are EPIC! Getting best results at CFG 3.5-4 at 15 steps.
It seems like a consistency model LoRA (my guess is on DMD2) was merged in at a pretty high weight, hence why only stochastic samplers (ancestral / SDE / Restart) work.
Try anything other than those and you'll get nearly unusable images - overexposure effect, really bad anatomy, etc.
Can you ELI5? Not arguing but just messing with it, I get lots of awesome images.
@The_AI_Will_Get_You Try using say DPM++ 2M sampler and you'll see what I mean by "nearly unusable images." You can also look on civit the images generated by this model - notice that ALL image that look nice used Euler A, while images with numerous bright color patches don't use Euler A
Specifically, I suspect that DMD2 was merged into it. You can think of DMD2 as a LoRA like Hyper, enabling very low step generation while making the result more noisy, which you combat by using a stochastic sampler, which is any sampler that additionally injects more noise during sampling, like Euler A.
As a rule of thumb, any sampler with Ancestral (A) or SDE in the name is stochastic, and so is Restart, seeds_2 and seeds_3. Things like DPM++ 2M is not stochastic, and when you use it the artifacts are obvious.
@11yu Probably yes, since i merged in with VXP-XL Illustrious which contains accelerator LoRA like DMD2 or Hyper, so other samplers won't work mostly.
Well with credit cards disabled, couldn't buy buzz so on-site generation will stop until restored... 😞
very fun model!
Flexible and convenient model, understands well the detailed description of a certain object.
Very pretty model but it's a massive pain in the ass and very inconsistent with characters. While some characters will be portrayed amazingly, unfortunately a majority have a very hard loli bias.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.