Version 3 CelestReal
Back to the drawing board with a change to the base models.
AniMayhem -Pale rider - True Grit version
Ultra Realistic v2- by Stable Yogi
Hyphoria Real [Illu] v5 - by ecaj
This is a swing back towards the semi-realistic and much closer to my original vision for this merge.
Style Highlights:
- Default look is a soft semi-realism that works across portraits, scenes, and NSFW
- Handles dynamic action poses, intimate compositions, and cinematic lighting without distortion
- Excels in detailed anatomy and smooth skin rendering
- character recognition is good, but some characters don't lend themselves to semi-realistic or photorealistic images without strange outcomes.
Style Tuning Tips:
- Add "anime" and/or "anime style" to re-enforce a more anime style.
- Add "realistic" and/or "photorealistic" to swing more towards lifelike results.
- leave both out for the default look.
Recommended Defaults:
- Steps: 25–40
- CFG: 4–6
- Sampler: DPM++ 2M Karras or Euler A
- Clip Skip: 2
- Resolution: Portrait 896×1120 or scenic 1024×1536
Upscale or face fix for mid distance and further. Most of the showcase images were created with a basic workflow and no upscale. Look at those for my personal settings. Things only get better with extended workflows!
Experiment and share!
As always - respect the original model makers licensing and the Civit TOS.
_____________________________________________________________________________________
CelestReal
Anime ↔ Realism Fusion (JANKU v4 + Babes Illustrious V4 FP16) + LoRA Enhancements
A finely tuned fusion that captures the charm of anime and the depth of realism in one versatile model.
---
Base Model Merge:
• JANKU v4.0 – “NoobAI EPS + RouWei Illustrious XL” by janxd
• Babes Illustrious V4 FP16 by Stable Yogi
LoRA Enhancements:
• TrendCraft – The People’s Style Detailer (Noob/Illustrious Base) by janxd
• Dramatic Lighting Slider IL by klaabu
This merge adheres to the original licensing and usage terms of all source models and LoRAs.
YOU SHOULD TOO.
Style Highlights:
- Default look is a soft semi-realism that works across portraits, scenes, and NSFW
- Handles dynamic action poses, intimate compositions, and cinematic lighting without distortion
- Excels in detailed anatomy and smooth skin rendering
Style Tuning Tip:
- Add "anime" and/or "anime style" to your prompt for a more stylized look
- Add "realistic" and/or "photorealistic" to swing toward lifelike results
Recommended Defaults:
- Steps: 25–30
- CFG: 4–5
- Sampler: DPM++ 2M Karras or Euler A
- Clip Skip: 2
- Resolution: Portrait 896×1120 or scenic 1024×1536
---
This is my first attempt at a merge and I'm pretty happy with how it turned out. Hope to see some outputs from all the talented people on here.
Description
A swing back towards the semi realistic with a complete rework.
Better results without additional loras or embeddings.
Adding photorealistic or realistic to your prompt will show drastic differences in the results.
More consistent with correct anatomy, hands tend to be more reliable.
Recommended Defaults:
- Steps: 25–40
- CFG: 4–6
- Sampler: DPM++ 2M Karras or Euler A
- Clip Skip: 2
- Resolution: Portrait 896×1120 or scenic 1024×1536
FAQ
Comments (19)
Sad news — the anime style is gone
It's not gone- it is more semi realistic however. You can always add Anime or Anime style to your prompts to swing it more that way.
for instance...
https://civitai.com/images/115853951
https://civitai.com/images/115856085
https://civitai.com/images/115853989
But if this isn't your preferred style, maybe check out -
https://civitai.com/models/1984400/animayhem-pale-rider
A really great checkpoint that's easy to work with and produces really beautiful pictures.
Whistler, a big thank you and much praise for your latest Checkpoint. You can really take great AI pictures with it. ^^
Does it generate men? For some reason, 1boy doesn't work and it always generates girls.
Yes it does, however it is going to lean towards women.
I'll post some examples.
hello,your model is very good!
but usually it created 2 girls even i typed 1girl,solo,how can i avoid that?
Without seeing what else you are prompting, it's hard to say. I don't have that issue.
Are you using widescreen format? I have this problem with almost every model, as I like to take widescreen photos. The AI tends to fill the entire frame. You can avoid this by trying to take close-up shots.
@lovingphi yes,thank you
@Whistler_ai i see,it maybe my problem
I like this model (v3) but it's constantly putting shadows across the subjects no matter how I prompt against it. Something like 8 out of 10 images are like this and it's frustrating.
Feel free to reach out. Lighting and shadows are strong in this so you will need to prompt specifics.
Anyone having problems with hands?
I have problems too. Too many fingers tend to appear, or two fingers merge together. I also have particular trouble when generated hands are supposed to hold an object, like a cup. Then the object and fingers merge, or the object isn't positioned correctly in the hands.
This is an excellent checkpoint that responds very straightforwardly to the prompt!
Despite the problems with animating hands, this is currently my absolute favorite model. It works very well with my LoRa.🙏❤️
100/10!
New favorite checkpoint.
This checkpoint is absolutely phenomenal. I used it in a photorealistic workflow, and I was genuinely blown away.
The background and lighting are simply next-level. Even with minimal or very simple environment prompts, it creates incredibly rich, atmospheric, and outstanding scenes that feel completely natural and cinematic. The way it handles complex lighting interactions and environmental details without needing heavy prompt engineering is pure magic.
On top of that, the textures are insanely fine and detailed. It brings ultra-realistic 3DCG characters to life in the most vivid and believable way I've seen in a long time. Skin, fabrics, hair, and subtle surface details all render with such exquisite micro-detail that it honestly feels like looking at a high-end production render.
This is easily one of the best checkpoints I've used this year. Outstanding work!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.