Socarealism XL 4.0
4.0 Update - Rebuilt with updated versions of checkpoints. Addition of a little bit of Yamer's Realistic NSFW and Rafaello into the mix. Major improvements in this version: does a much better job of a range of lighting/moods. Significantly improved skin/face texture. Better variety in faces/people.
3.0 Update - Rebuilt again from the ground up (as I will do with every x.0 update) using updated versions of the checkpoints. Added Hephaistos DPO at the end as a finisher, which really helps prompt responsiveness. Changing naming conventions to cut down on filename size for folks (Socamerge Realism XL becomes SocaRealism XL).
2.0 Update - Rebuilt from the ground up merging in newer versions of checkpoints. Added Nightvision into the mix. I'm pretty happy with the status of this checkpoint merge, so unless a new realism checkpoint comes out that really wows me, future updates will mostly be remerges to bring in new details / enhancements from updated versions of what is already merged in.
1.5 Update - Low strength merge of some updated checkpoints, to get the changes from their new editions mixed in.
This model is a generalist checkpoint merge that will produce photorealistic images, art, anime, stylized output, and everything in between. Preview images are mostly of hawt photorealistic ladies because I have an unashamed preference for such, but it is also quite capable of producing men and all folks in between, backgrounds, different styles, etc. It is quite flexible and works well with a wide range of Loras. It is very good at NSFW if you prompt for it:
NSFW Prompting:
- (Nude naked NSFW) in positive prompt = very naked people.
- No (Nude naked NSFW) anywhere = a bit on the skimpier side for clothing.
- (Nude naked NSFW) in negative prompt = more conservative clothing, can reinforce descriptions of clothes that cover-up more.
It works very well with DPM++ 2M SDE Heun Karras and DPM++ 2M SDE Heun Exponential sampling methods, but should work just fine with most others. I've tested it with a wide range of samplers and steps. All my uploaded images were generated with DPM++ 2M SDE Heun Karras and low strength ADetailer (often only Eye ADetailer to get the sclera/pupils more defined).
This checkpoint does have a VAE baked in, as that was needed from some of the checkpoints I merged in, so be aware of that if you do further merges (just make sure to bake a vae in as you merge). Automatic VAE selection will work just fine when you generate images with this checkpoint.
Huge thanks to Afroman4Peace, KandooAI, CafeFlesh, and Fodda, without whom this wouldn't exist - and whose work provided the inspiration for the original merge.
ADetailer: search for ADetailer in the WebUI Extensions panel or go to: https://github.com/Bing-su/adetailer
Models merged in:
Juggernaut XL
https://civarchive.com/models/133005/juggernaut-xl
DEMON CORE
https://civarchive.com/models/155977/demon-core-sfwandnsfw
Pandora's Box
https://civarchive.com/models/143067/pandoras-box-nsfw
Fodda XL
https://civarchive.com/models/130697/foddaxl-photorealism-for-sdxl
7 of 9
https://civarchive.com/models/144313/7-of-9-scenic-edition
New Reality
https://civarchive.com/models/161068/lnhznewrealityxlv10
RealVisXL
https://civarchive.com/models/139562/realvisxl-v20
Hephaistos NextGen DPO
https://civarchive.com/models/241445/hephaistos-nextgen-dpo-swfandnsfw
Yamer's Realistic
https://civarchive.com/models/127923?modelVersionId=299716
Photonium
https://civarchive.com/models/268578/photonium?modelVersionId=302809
Description
Small update focused on lighting, face variety, and hands.
FAQ
Comments (5)
Lets see what folks make with this update! It's good at landscapes and abstract as well!
Nice work bro, One of the top notch realistic SDXL model at the moment. Really appreciate your efforts.
I am Eric, I run a Gen AI startup leveraging SDXL Loras and want to suggest a collaborative research opportunities making 'Midjourney for human portraits'
Let's jump on a quick call/chat and I would cover detail + opportunities we could make together.
many credits, reputation and compensation is assured.
Find me on the contact below:
- Discord : eric_sdxl
- Email: [email protected]
As I am not the rights holder to the Loras/Checkpoints merged into this model, I'm not going to be able to sign up for this. I list what I used to make this checkpoint in the description - those authors should be contacted rather than myself.
@Socalista thanks for honest feeback. but are you model trainer yourself as well?
@ericmaeng I am, but havn't yet released the model I'm training - takes a lot more time and computing effort to get it right compared to merges.