Join me on Patreon for exclusive perks and early access to unique resources.
To discuss custom LoRa's or models, feel free to connect on Discord.
Meet ULTRA_REALISTIC_BY_STABLE_YOGI
Why a new model?
I fused the strengths of my two flagships—Realism_By_Stable_Yogi (hyper-detailed textures) and Babes By Stable Yogi (stylistic “babes” aesthetic)—to remove the hassle of juggling checkpoints.What’s in it for you?
Single download, double power: One drop of "Ultra Realistic by stable yogi gives you the photoreal skin, fabric and lighting of Realism plus the flattering poses, proportions and vibe of Babes.
Cleaner prompts: No more switching models or stacking LoRAs; just load, prompt, and render.
Consistent outputs: Both styles are blended and re-balanced so results stay sharp and coherent across various seeds and resolutions.
Bottom line
ULTRA_REALISTIC_BY_STABLE_YOGI lets you tap the best of both worlds—studio-grade realism and the polished “babes” look—without extra setup. Grab one checkpoint, create more.
Description
Pro Version of most of my models are available on My Patreon
This version features the latest merge of the Realism and Babes models, plus a few extra tweaks.
Sampler Euler a, DPM2 a, DPM++ SDE, DPM++ SDE Karras, DPM++ 2M SDE-- SGM Uniform,
CFG Minimum 4+ maximum upto 7
Steps Minimum 18 , Recommended Steps 27
Resolutions: All SDXL Resolutions
Adetailer (Required)
High res Fix (Optional)
Denoise - 0.30
Minimum hires steps 5+
Minimum Upscale 1.5
Upscaler 4x-UltraSharp
Keep the VAE at Auto ( VAE Included )
Negative Prompts
(Download this negative embedding for best results) Stable_Yogis_PDXL_Negatives
Positive Prompts
(Download this positive embedding for best results) Stable_Yogis_PDXL_Positives
FAQ
Comments (38)
The V2 DMD2, I never thought 4 steps would work, but it does. Ultra-realistic in 4 steps. I use 6 steps because I feel the understanding is better. Thanks Yogi, it's amazing. And no Problems with Finger and Nails out of the box... wow
its 2 fast, cant take it all in
does anyone know why these nodes are missing?
UltimateSDUpscale(In group node 'workflow>Tile Upscale')
workflow>Tile Upscale
LoraStackLoader_PoP
ImageSmartSharpen+
thats what it says when i copy the work load into comfyui
@ghostbirdlary images are generated in A1111, comfy ui may not grab all nodes. you may need to add manually.
where would i get the nodes to add manually?
@ghostbirdlary I'm using the ComfyUI App (the one that doesn't have to use a browser). Anyway, if you go to the Manager there is a button that says "Install Missing Custom Nodes" and it has an 80% success rate. After that, if it doesn't find them all, you can go to "Manager" and then "Custom Nodes Manager" and search manually for whatever is still missing. Last ditch efffort, Google searches will generally lead you to whatever still hasn't been found.
@WillieF where is the version that doesnt use a browser? when i launch comfy ui it always opens in my browser with a local ip
@ghostbirdlary https://www.comfy.org/download
Just download the one for Windows or Mac, not the GitHub version. I like it much more. It seems faster, and it's definitely more stable and everything just seems easier to use.
OK, ok so hear me out, I've been having this issue since forever but the upscaler actually works. So if it brings you any piece of mind, you should try getting the github link and updating it that way, that's how it worked for me. Do keep in mind that I wrote this while drunk so if you have any inquiries do direct them to me and I may answer them timely.
When I try a positive prompt like this one: [Stable_Yogis_PDXL_Positives,(random strong wrestler, female powerhouse, masked luchador, unmasked champion, heavyweight brute, agile high-flyer), professional wrestling portrait, upper body, 85mm lens, neutral studio backdrop, depth of field, (high detail skin:1.1), slight sweat on skin, wrestling gear, tights, optional championship belt, optional mask, dynamic pose, cinematic softbox lighting, rim lighting, RAW photo, photo-realistic, ultra-sharp focus, studio photography, symmetrical composition, detailed muscles, skin pores, (AOI:1.2), intricate details, dynamic lighting, shallow depth of field, photorealistic HDR] with this negative: [Stable_Yogis_PDXL_Negatives-neg, (lowres, bad anatomy, bad hands, missing fingers, extra limbs, blurry, watermark, text, jpeg artifacts, deformed face, mutated, cross-eye, lopsided eyes, cartoon, painting, drawing, illustration, 3d render, cgi, plastic skin, duplicate face, long neck, fused limbs, (nsfw:1.4), out of frame, bad proportions, ugly)]
Sampling method: DPM++ 2M SDE
Many times it generates a naked person.
This is what i created https://civitai.com/posts/19407784
refine your prompt.
@Stable_Yogi Thanks. That works
having minor trouble with this one, a lot of artifacting in hair and skin regardless of cfg or sampler. not really sure what's wrong.
nevermind, I had tried merging it with an inpaint model and there was some internal mismatch.
So what's the difference between FP16 and BF16?
Use fp16 for precision,
Use bf16 for stability and broader hardware support.
Logioges could you be more specific so a newbie can understand about precision/stability you are talking about? Thanks!
Not bad from what I’ve experienced so far 👌
By far the best checkpoint I've used. Better than SY's Realism or Babes, and better than CyberRealistic which is my current #2 Pony. Would love to see it permanently available. Please up vote it!
Amazing results, great checkpoint!
v 2 fp16 is amazing, thanks for your work. Keep it up!
Will there be a newer version than SD 1.5? I'm asking because only SD 1.5 models can be run on mobile NPU for now, and this model is older than the latest versions of the other two models.
Sorry bro, No plans for SD1.5 for now.
Which is the best UI to run this checkpoint? Is there any quality difference between fooocus and comfy?
Use Forge neo, its better
@Stable_Yogi thank you. will you make one of your amazing checkpoints for z-image?
@MrPurridge https://civitai.com/models/2221503 Its already done, check here
@Stable_Yogi tks!
@Stable_Yogi hey Yogi, how do I use your negatives and positives embbedings on forge properly? tks in advance!
@MrPurridge Download and save them to embedding folder, restart UI
open the Embeddings tab, You see it there.
then use that in prompt. YT can help, not that difficult.
@Stable_Yogi thank you. yeah I know but since it's your own embdgs I though you could answer here quicker. also I've sent you a message on discord asking about your patreon tiers and you didn't answer there, so I don't know exactly where you answer.
@MrPurridge Responded all messages on Discord. Check there pls.
@Stable_Yogi nope, nothing.
@MrPurridge check
@Stable_Yogi done
This model certainly seems to combine aspects of the realism and babes models, exactly as the creator describes. It has also inherited a trait of the babes model, which is glamorizing everything. The original realism model can be prompted with a recognizable fictional character and it seems to try to recreate that character, but "real." Hair styles will be similar, facial features will be present, etc. Adding tags for realism (realistic, photorealistic, etc) changes this behavior even in the realism model and it begins to glamorize actors, bringing them closer to the babes model.
The babes model doesn't need these tags to do this. It starts out with actors being generated as supermodel versions of themselves, if they're recognizable at all, and realism tags don't change much or just tighten up features here or there.
The point is, this model is just like the babes model in that regard. There is less "true to life" anime character and more "possibly the character, if they were real, and also a supermodel" here.
what ControlNet do we use with this model?
SDXL