Get the pro version of my checkpoints and workflows on my Patreon.
Onsite generations are permanently available on these models:
👉 Realism_Illustrious_By_Stable_Yogi : https://civarchive.com/models/974693?modelVersionId=2091367
Read "About this version" Section for Model details and settings.
Important Usage Tips to get similar results like my images.
Add Realism_Illustrious_Negative_Embedding at the beginning of your negative prompt.
Add Realism_Illustrious_Positive_Embedding at the beginning of your prompt.
Join me on my Patreon for exclusive perks and early access to unique resources.
To discuss custom Lora's or models, feel free to connect on Discord.
đź‘Ť Like this model to keep me motivated and inspired to create more!
đź’¬ Drop a comment and let me know what you'd love to see next.
🌟 Review this model to help me improve and make even better creations.
đź”” Hit that notification bell to stay updated with my latest models and updates!
Description
What’s new in this release (and how you can tweak your prompts):
Native Insta‑style portrait framing
– True 4:5 aspect ratio, optimized for 1080 × 1350 postsBalanced exposure & natural lighting
– Soft yet precise highlights/shadows—no blown‑out whites or deep, harsh blacksUltra‑realistic skin detail
– Light, natural pores and freckles (((freckles))) with realistic catch‑lights (((catch-lights)))Flexible depth of field
– Blurry background mode for “portrait” vibes or full‑scene clarity for street‑style shoots (just add(((add depth_of_field)))to your prompt)Consistent color grading
– True‑to‑life hues with a touch of warmth for a cohesive gridSFW & NSFW output
– Feels like a pro mirrorless shot—ready to drop into your feedAdd Your LoRAs & See the Magic
All images above are raw generations—no post‑processing—so you can judge the model’s true capabilities.Recommended Samplers
LCM
EULER A
Optimal Settings
Steps: 6 or 8
CFG Scale: 1.01 or 1.1 (1.1 makes sure that your negative prompts work)
Clip Skip: 2
Resolution: Any SDXL size; for Insta‑ready posts use 896 × 1120 px
High‑Res Fix: 1.206× upscale with the 4× ClearRealityV1 upscaler (9 steps, denoise 0.1)
Looking for feedback:
We’ve aimed to strike the right balance between “just‑snapped” realism and polished, editorial lighting. If you spot anything that could be smoother (skin tones, background blur, color warmth, etc.), we’d love to hear what you think!
Join us on Patreon here.
FAQ
Comments (71)
I think v3.0 dmd2 is the best
Even better than the newer versions?
v5.0 dmd2 is pretty good too, much better than 4 and 4.5
Super Noob question, How well SDXL characters models perform on Illustrious realistic checkpoints? or they just don't work?
that depends on Lora and the checkpoint, you can test and check yourself.
@Stable_Yogi Oh I see, at least there's a chance! thanks from the info :)
Bit of a newbie question. I can get amazing beautiful results with this checkpoint (and associated LoRAs). Hair styles and colors work, women it generates are gorgeous, etc.
But what prompts for pose and view angle do I use for an image directly above of the subject lying on her back in bed?
I have never had any luck with this view and pose. I always wind up sort of eye level from the front and at best she might be leaning back on the bed, never laying flat and directly above.
Any prompt help would be appreciated, thanks!
gemini is trained on civitai ask it
Did you try aerial shot?
laying down, on back, from above
There are working better, thanks.
Guys, I need help, I generated a model using a simple workflow of one load checkpoint (the one from here), 2 clip text encoders, one positive and one negative, ksampler with all the settings, empty latent image, the vae decode and the save image, what do I use to recreate the same model face with the same body features or almost the same using the same pre-trained checkpoint? I a total newbie, I did everything with Chatgpt's help, I kinda learned some things but it is still too complicated for me, I saw that there's some models that you can use for generating the images with the same face but I don't know how to make another workflow so that I can use the same face and still use the same checkpoint, I really appreciate your help, thank you.
I am a newbie like you. Getting the same exact face as someone else could be problematic because you'll need everything they used plus you need all of their settings.
Its like this... let's say you see a car you like and you want one just like it. If you really want one just like then you'll need all the parts they used plus you will need the equipment they used to make the car. Each thing they had to make the car that you don't have will change what your car will look like. Maybe not by much but it will change.
If you want to be in the ball park you definitely need the same checkpoint, Lora's, clip , and vae. The actual prompt would be nice. You'll need all the KS Sampler information. I have been at this for awhile off an on. I finally got my private server setup a few days ago. It's a learning curve.
Watch tutorials and of course ask for help. I don't know if I have been helpful or not but this is a mission of frustration, but I am willing to learn.
One last thing just to let you know how sensitive some of these settings and then are, I once processed a picture at 25 steps then up the steps to 80 just to see what would happen and I got a completely different picture. All other settings were identical, seed and everything.
Out of all the XYZ plots I've done, with/without "quality" modifiers (ie: masterpiece, best quality, 4k, Photorealistic, etc), with/without realism lora's, with/without yours, easypos, and cyberreal embeddings, in my opinion version 4.5 takes the cake. V5 seems too.. undetailed? unclear? I'm not sure how to describe it. Kindof like having a texture stretched over a model if that makes sense. Not enough detail, or unfinished, out of focus. The prompting is good on both, they provide pretty much the same output, but different levels of detail. With more clear, and overall better detail in V4.5. Or maybe V5 requires different prompting that I didn't attempt.
FWIW I'll add to this that I've found that training LoRAs with v4.5 produces better results than v5
Beginner question... Using Comfyui, I have added the checkpoint, I added the positive and negative embeddings (and restarted the app). When I drag/drop one of the sample images and hit render...
1) I have to change the checkpoint from "Realism By Stable Yogi Illustrious V5 FP16" to "realismlllustriousBy_v50FP16.safetensors"
2) After doing this, it renders, but the image is full of artifacts.
What am I doing wrong?
Confirm first that you are following these suggestions please
Recommended Samplers:
DPM++ 2M SDE, Euler A, DPM SDE,
Optimal Settings:
Steps: 27+
CFG Scale: 5
Clip Skip: 2
Resolution: Any SDXL resolution, 896*1152
I had the same issues. Removing the Negative/Positive embedding keywords fixed it. I imagine its because I didn't have the embedding loaded.
Very very very good model for realism. v5 FP16 rocks. Thank you for sharing it.
Does this works with illustrious character Loras?
Yes
Hope this makes it into next week's selection of Models, it's one of my favourites.
Sure i will add this, also please bid on the models page.
Where recommended parameters?
check the "About this model" section.
What is the difference between DMD2 and DMD2_ALT models? I haven't seen anything about it inside the "About this version" stuff.
Curious about this too. Since I just got wind of DMD :D
DMD2 is a lightning variant , uses few steps and Low CFG, like 6 Steps and 1 CFG works too. (Quick images)
DMD2_ALT is a lightning variant too, but its CFG can be used from 1 to 7, and steps can be used from 6 to 30, and all samplers work with this model now.
This helps with detailed images and higher CFG ensures prompt adherence.
How to use ?
When you increase CFG you increase the steps.
For each 1 CFG you add 4 steps.
CFG 1 = 6 Steps
CFG 2 = 10
CFG3 = 14 and so on
Hello, I made a merge using this model, can I share it if I credit you in the description?
Sure, on your model description page add the credit with the link to this model.
Let me know when you do that.
thanks
What's the difference between Realism Illustrious and Babes Illustrious? I can't decide which to choose.
Realism = Overall Realism Focused Images
Babes = Focus on Influencers type Images and textures
Ultra Realistic = Mix of Realism and Babes
Thank you very much for the explanation. Amazing models btw
I started using this model 2 days ago, (just a beginner) but this has become my first and favorite model after trying out a bunch of other popular ones. Is there any chance you could bring this back as one of the selection models here? Much appreciated.
Sure will do that, this week.
Can you include this in the next selection please? I bid some buzz in an attempt to give it a push.
yes i will do my best.
Simple prompts, Amazing results. The force is indeed strong with this one.
When I load your workflow (from image) to comfy, the image only has artifacts. I have downloaded the embeddings aswell...
Image sin gallery are made in A1111, comfy wont give you same results.
Newbie question: Can I use this checkpoint to make visual novels or YouTube videos without any legal issues? I'll credit you, of course.
Sure, you can use if you credit.
Thank you. 🙏
Beginner AI image creator here. When training a LoRA to be used on this checkpoint, should I train it on this checkpoint specifically or would training it on generic illustrious suffice?
Additionally, is there anything you'd recommend regarding the LoRA training to make it better for this model? Thanks!
Train on this model. Nothing special required.
@Stable_Yogi Thank you sir!
Each version is getting better, and more and more flexible! I've got all of them and run the same prompt through each, and each version gets noticeably superior! Good work! One pose that it seems to be unable to do (other illustrious models struggle too, but I can usually get them to do it with some work) is legs straight up in the air. No matter what I try, the feet stay firmly pointed down. Other than that, it has done every pose, style, and LoRA I've thrown at V5! Mega kudos! Hope you'll keep improving it! :)
Totally new at this. I don't see any recommended settings, unless I'm just missing it. Also, can you point to a tutorial how to add your Positive, and Negative prompts please and thank you.
Please check the "About this version" section for all details on this model.
Just add embedding in the Positive and negatives in A1111 or Forge. easy.
@Stable_Yogi Thanks so much for the reply. I'm using ComfyUI. I dragged one of your sample images into ComfyUI and it mostly worked besides some artifacting. Changing the sampler to 'euler_ancestral' fixed the artifacts. Thanks again.
❤❤❤❤❤❤❤
Having trouble doing black women with this model, already tried african_woman, african_american, dark_skinned, everything but they come out looking more mexican than black, anyone else?
use a dark skin lora like https://civitai.com/models/466467/skin-color-slider-pony-illustrious
also danbooru tags are your friend this is an illustrious model so it's heavily trained on danbooru tags so the correct tag would be dark_skin or very_dark_skin, although you might want to try removing the underlines so see which one works best as some work better with than without and vise versa, and you might want to consider removing underlines when it's a matter of degree, so "gigantic breasts", "huge ass", "puffy lips" etc will work better because it will use the natural language interpretation as well as the danbooru tags.
the term "ebony" has worked for me on many models
I don't understand why the author don't give a detail instruction on the new released checkpoints? Just wait us to ask?
"Read "About this version" Section for Model details and settings."
@SaburoDio Not "detail" enough, e.g. Additional VAE needed? And I find some of these checkpoints very difficult to work with Lora CLip strength & Lora Strength setting. How to get around it?
@Tsinghan Usually people mention if a VAE is needed and I have not once had that issue.
"Recommended Samplers:
DPM++ 2M SDE, Euler A, DPM SDE,
Optimal Settings:
Steps: 27+
CFG Scale: 5
Clip Skip: 2
Resolution: Any SDXL resolution, 896*1152"
And regarding Loras: That's up to you or whoever made the Lora?
It seems like you just generally lack understanding and knowledge on it and you need to learn more yourself instead of being rude.
@Tsinghan VAE is mostly included in XL based checkpoints, so no need to mention that.
If you could simple check the images in gallery, all meta and code is there, just need to look carefully.
Lora strength is not fixed, you adapt that based on your requirement and condition of the image. No need to get around, need to learn and understand the basics.
@SaburoDio Sometimes there is no "About this version" section, like with v5.5_DMD2 for example
@bovinespatula listen buddy. No one can help you guys to use your brains.
Three 5.5 versions and two have the text. The other one obviously literally only adds DMD2.
You lack basic knowledge of everything and blame the uploader of FREE (!!!) models for you to use because you are too lazy to expand your very limited knowledge.
I guess this is all the I-pad kids as grown-ups?
@SaburoDio I just merely remarked that some uploads don't come with info. You should take your own advice and stop being rude.
@bovinespatula I apologize. There are just so many "toddlers" on this page constantly whining about minor things on free checkpoints.
Do we need to download the embeddings safetensors files or are they included in this checkpoint ?
Dreckscheckpoint!
v15 bf16 looks amazing, how is it not the top slot? Genuinely curious
Realistic VS Ultra Realistic for Illustrious & SDXL, which is better?
Whatever works best for you, is the best. I craft all models with much attention.
Any planned updates for this or have you moved on to other things?
All models are alive and planned, work in progress, a bit slower.
Sincere apologies.
But stay tuned, all models will be updated soon.
thanks
@Stable_Yogi Hi, sorry for the inconvenience.
Could you tell me which one is more realistic and which one understands Danbooru tags better between this model and Babes Illustrious by Stable Yogi?
Also, which one works better with prompts for characters from anime or TV series?