epiCRealism - CivArchive (CivitAI Archive)

epiCRealism - pure Evolution V3

NSFW

Natural Sin Final and last of epiCRealism

~~Since SDXL is right around the corner~~, let's say it is the final version for now since I put a lot effort into it and probably cannot do much more.

I tried to refine the understanding of the Prompts, Hands and of course the Realism.
Let's see what you guys can do with it.

Thanks to @drawaline for the in-depth review, so i'd like to give some advices to use this model.
[expand to see Advices]

Advices

use simple prompts
no need to use keywords like "masterpiece, best quality, 8k, intricate, high detail" or "(extremely detailed face), (extremely detailed hands), (extremely detailed hair)" since it doesn't produce appreciable change
use simple negatives or small negative embeddings. gives most realistic look (check samples to get an idea of negatives i used)
add "asian, chinese" to negative if you're looking for ethnicities other than Asian
Light, shadows, and details are excellent without extra keywords
If you're looking for a natural effect, avoid "cinematic"
avoid using "1girl" since it pushes things to render/anime style
to much description of the face will turn out bad mostly
for a more fantasy like output use 2M Karras Sampler
no extra noise-offset needed, but u can if you like to 😉

How to use?

Prompt: simple explanation of the image (try first without extra keywords)
Negative: "cartoon, painting, illustration, (worst quality, low quality, normal quality:2)"
Steps: >20 (if image has errors or artefacts use higher Steps)
CFG Scale: 5 (higher config scale can lose realism, depends on prompt, sampler and Steps)
Sampler: Any Sampler (SDE, DPM-Sampler will result in more realism)
Size: 512x768 or 768x512
Hires upscaler: 4x_NMKD-Superscale-SP_178000_G (Denoising: 0.35, Upscale: 2x)

Useful Extensions

!After Detailer | ControlNet | Agent Scheduler | Ultimate SD Upscale

⁉ No VAE needed but it is better to use one for more vibrant colors

⭐ Feel free to leave Reviews and Samples - and always have fun creating ❤

Description

Lets say we reached the final Version, tried to improve male understanding, hands and of course photorealism

FAQ

Comments (53)

flubberducky123Jun 27, 2023· 2 reactions

CivitAI

Could you make a new inpainting model for V3? Your models are some of the best for realism, absolutely love them.

epinikion

Author

Jun 27, 2023· 4 reactions

Sure i will do. Pls be patient a bit

epinikion

Author

Jun 27, 2023· 3 reactions

Got you covered! It's there :)

futurafreeJun 27, 2023· 2 reactions

@epinikion You're the goat man

CreativeDudeJun 27, 2023· 3 reactions

CivitAI

Incredible model! I feel like this is the closest to realism I've seen in a model. Nice job!

epinikion

Author

Jun 27, 2023· 1 reaction

Thank You! ❤

SLOPEJun 27, 2023· 7 reactions

CivitAI

Definitely the best realism model I've seen so far. It seems like SD 1.5 is approaching it's maximum potential here

1751002Jun 27, 2023

CivitAI

Great Work and thank you for sharing! Wich female Character Embedding (LoRa or Textual) is recommended fpr this Model?

RomazeoJun 27, 2023

CivitAI

Can you say that this model is going in a way of better hands and fingers wrapping? And more logic in details?

zx3oJun 27, 2023· 18 reactions

CivitAI

I think that in order to get even closer to realism, we should not strive 100% for the ideal. Realism is given by shortcomings, skin defects, small disproportions, etc... but these are just my thoughts. ;)

OttoDAIJun 28, 2023· 3 reactions

CivitAI

Anyone got mirrors for 4x_NMKD-Superscale-SP_178000_G? The one linked is public transfer limited exceeded

lostdogJun 28, 2023· 4 reactions

here https://nmkd.de/shared/?dir=ESRGAN/Models/Realistic%2C%20Multipurpose

OttoDAIJun 28, 2023

@lostdog thank you kindly!

rpgamezzzJul 1, 2023

@OttodAI where do i put the file to?

epinikion

Author

Jul 1, 2023· 2 reactions

@rpgamezzz Models/Esgran folder

rpgamezzzJul 1, 2023

@epinikion thanks

andersJun 30, 2023· 8 reactions

CivitAI

The lighting on V3 is so satisfying. This is my favorite model for realism. Thanks for sharing it along with the embeddings.

epinikion

Author

Jun 30, 2023

Thank you, glad you like it ☺️

unickJul 3, 2023

CivitAI

There is something weird happening in combination with controlnet: it fails for the "reference" control type. Even more so with "ControlNet is more important". The output becomes almost completely black. It happens only with this model (and mixes using this model). I wonder what is happening.

unickJul 3, 2023

Above comment is for V3. I just tried V1, that also has this problem of images getting darker with controlnet "reference" but not as bad. The image does not turn completely black, just very dark.

epinikion

Author

Jul 4, 2023

@unick I don’t have this happen to me. Probably there is something messed up with your installation?

AIbyIndigoJul 4, 2023· 1 reaction

CivitAI

Thank you for your work!

1720435Jul 4, 2023

CivitAI

Anyone got tips how much Unet training steps per instance image to train with his model? Using fast dreambooth to train my data set, and usually i use 100 steps/instance image but somehow the face wont use my instance image. sry for the bad english.

247103Jul 4, 2023

CivitAI

Is kkw-ph1 a lora?

epinikion

Author

Jul 5, 2023

Yes it is linked in suggested ressources „easy photorealism“

Stable_ConfuzionJul 5, 2023

kkw-ph1.bin is a TI

epinikion

Author

Jul 5, 2023· 1 reaction

Right, it is a TI 😄

grimreaperalphax6915Jul 5, 2023· 2 reactions

CivitAI

This is Godzilla of rendering models,nothing can even come close to it for now,it's so realistic that's actually scary but like the author said keep negative at basic.

Best way is to just use author negative prompt,because model is so amazing that would never render bad images.

Main thing is CFG scale needs to be left at 5 like author said because difference between numbers is massive,just switch it to 5 and forget about it.

Amazing job!

epinikion

Author

Jul 5, 2023· 1 reaction

Haha thank you! ❤️

DowntroddenJul 5, 2023

I've been having good results below CFG 5. With 1 colors blend together nicely if thats the vibe your going for. 2-4 is sweet spot for me honestly

Boxer766Jul 6, 2023

Whats the best sampler for this model?

DowntroddenJul 6, 2023· 1 reaction

@Boxer766 I quite like using DPM++ 2M SDE Karras but i havent tested the others too much.

grimreaperalphax6915Jul 6, 2023· 1 reaction

@Boxer766 DPM++2M SDE Karras working the best for photorealism.

grimreaperalphax6915Jul 6, 2023· 1 reaction

@epinikion No thank you!,i just forget about other models because this beast can do anything: beautiful nudes,gothic fantasy,SF,basically whatever you want with right prompts.And i noticed model is super sensitive on prompts so you can just name the words with weights and model would just explode,never seen such quality or precision in any model.

You are genius and keep up with good work!

crnclashers868Jul 6, 2023· 2 reactions

CivitAI

@epinikion I have accumulated 160k images from unsplash and 1 million images from 1x.com. I want to train a hyper realistic model based on sdxl, can you please collaborate with me? lets make this big, I have an investor who can help us with the computing power.

we can make this big

aimannJul 8, 2023

Hope this works out. Looking forward to it.

rebixJul 10, 2023· 1 reaction

do it pls

SYYNJul 6, 2023

CivitAI

why are my images coming out blurry and fuzzy? i see other peoples images and they look amazing and real?

ktiseos_nyxJul 7, 2023· 1 reaction

CivitAI

In order to lesson confusion imma go name my "EPIC MIX" with realism to something less confusion LOL <3 also your model is amazing, never stop creating <3 thank you for making amazing stuff

GrigPerJul 7, 2023

CivitAI

Why i get

AttributeError: 'AttentionBlock' object has no attribute 'to_q'

when trying to train Lora in Kohya Collab?

mienaitekiJul 8, 2023

CivitAI

I can't get this google collab to dreambooth train from this model: https://colab.research.google.com/github/sagiodev/stablediffusion_webui/blob/master/DreamBooth_Stable_Diffusion_SDA.ipynb#scrollTo=XU7NuMAA2drw

On huggingface, I don't see any official model there from you either, and none of the ones available seem to work. Their file structure is wrong, and they don't have any fp16.

Do you have any advice of how I could dreambooth train a model based on this? It's my favorite realism model by far!

futurafreeJul 8, 2023· 1 reaction

Just right click download button and "copy link" my brother

DowBrown979Jul 11, 2023

CivitAI

I use Easy Diffusion and I'm getting this error:

Error: Could not load the stable-diffusion model! Reason: Error(s) in loading state_dict for LatentDiffusion: size mismatch for model.diffusion_model.input_blocks.0.0.weight: copying a param with shape torch.Size([320, 9, 3, 3]) from checkpoint, the shape in current model is torch.Size([320, 4, 3, 3]).

Anyone knows hot to fix it?

thestanmanJul 12, 2023

CivitAI

I just have to ask, how are you able to train such high quality photorealistic images off of SD 1.5??? Did you use any trigger words, or...

whoaoh27719Jul 12, 2023

CivitAI

Sometimes while loading this version of Epic it will crash automatic 1111. Works fine the next time and the next few times then it will crash it again, anyone having this? im only having it with this model.

epinikion

Author

Jul 12, 2023

Do you get any error messages?

anidnkJul 13, 2023· 1 reaction

check the size of the files, i had problems with another model because the file was incomplete when i downloaded it, i downloaded it again and problem fixed

DevScoTechJul 16, 2023

CivitAI

Did you plan a new pix2pix version?

epinikion

Author

Jul 16, 2023

Not really, you can merge it yourself like inpainting model merge. Since ControlNet nobody really uses pix2pix anymore I guess.

vinpunk99Jul 16, 2023

CivitAI

How do I deal with the color cast on this model ?

Don_DudeJul 16, 2023

CivitAI