I try to pin images that stand out / i personally like every 1-3 weeks
So if i unpin yours doesnt mean that i dont like it anymore just trying to keep a fresh rotation of images without completly filling the start page with pins <3
Join the Discord for questions / sharing pictures :) : https://discord.gg/gcJqAKQ5Af
If you enjoy my models and want to support me, you can now support me on kofi
(Since my patreon got disabled like a year ago & buy me a coffee banned me 4.2026 )
š ļø Recommended Settings:
Resolution:
best results with832x1216//960x1440//1024x1024//1024x1536CFG Scale:
5ā8Sampling Method:
Euler aorDPM++ 2M KarrasSteps:
20ā50
Description
going back to the roots of v2
FAQ
Comments (25)
Nice
The most beautiful checkpoint!
It works great with almost any LORA.
Maybe I just got used to it, but I think that at least to me V.4 looks better than the new V.5
Id say V.4 looks overall better than V.2, and also better than the new V.5 version. V.4 is so far peak :D
i dont know if i fully agree i havent generated that much with v5 yet and didnt rly have time to generate alot with v4 either, it just focuses more on "realism" instead of anime style ish
After producing some images, I feel that v5 has improved in terms of realism, but it is not as good as v4 in terms of color and lighting,
My personal feeling, v4 is more balanced between color and realism.
so glad I found this. Results look amazing
I really want to love this checkpoint, but I just canāt bring myself to use it. The fingers just arenāt fingering, there are often too many if you use hand-specific prompts, or they end up as tiny nubs or warped shapes. On average I had to do around five generations of the same prompt just to get decent fingers.
And sometimes, though itās rare, if the character smiles with teeth, the teeth look painted on or just weird. Itās sad, because aside from these two major flaws, this is one of the most beautiful checkpoints Iāve ever used.
For reference, Iām generating at 1024Ć1536 with 1.5 Hires Fix (R-ESRGAN 4x+ Anime6B), 0.4 denoising, and 20 HF steps.
Iām also using 7 CFG, 30 steps, and Euler A.
My standard positive prompt was: masterpiece, best quality, newest, absurdres, highres, 8K, ultra-detailed,
Maybe Iām doing something wrong, but if I am, I honestly canāt spot it.
I experienced exactly the same problems.
Try adding these tags to your negative prompt:
missing thumb, deformed hands, malformed hands, extra fingers, fused fingers, wrong number of fingers
For teeth I just ended up mostly using 'closed mouth' with something like 'seductive smile'.
Just chiming in that these flaws are still present in version 6. Variety on male faces also leaves something to be desired, I either get the gigachad jaw or it switches to a female shape, with nothing in between.
@FrivolousBand10Ā some of theese "flaws" are just illustrious / ai generall problems, when it comes to faces u can try lower cfg scales to if u havent yet
@FrivolousBand10Ā but to be fair i dont rly train much on men
Can u fix the policey thing in pixai it's not letting generate the image
i dont use pixai
@Greed25 so i dont know how to fix that ig u have to write pixai or idk
@DesyncĀ i did massage that model description but nothing changed its probably need to upload again i only did generate one time after that it got restrictions
Ok, maybe I was wrong. Did some more testing with V.5 and it actually looks great. At first I thought overall lightning/shading was bit worse, but that could have been kinda randomness in pictures, depending on the enviroment like certain types of indoor rooms etc. This can be easily just my feeling, and somebody might notice it different, but Id say that faces look overall bit better in V.5 than on V.4
Just started generating AI artowork and this checkpoint is amazing! I look forward to creating more scenes with this.
Wish this was available on the website again.
v5,4,3 are avaiable for generation
I think this performs on par with Z Image Turbo. It's the first time I've seen consistently coherent results from an SD/SDXL model.
Unfortunately it is not strong on the photorealistic side. Perhaps a LoRA might fix that.
this isnt meant to make photorealistic stuff, its only supposed to make 2.5d anime. also stop comparing base model to a extensive finetune
Right, it's more anime-ish than _50 version.
unfortunately i can't create a solo man
Skill issue?
https://civitai.com/posts/25563079
Facial variety isn't that great, but it's not like some other checkpoints that bolt tits on everything that has long hair...
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.