This is a trained checkpoint for creating photorealistic hardcore group sex scenes, especially featuring the act of double penetration (dp).
Most consistent results with CFG 7 and 40+ steps using DPM++ SDE Karras. See example images for starting prompts.
For best quality recommend upscaling using 4x_Foolhardy_Remacri with 0.35-0.55 denoising
If you enjoyed this model, please consider leaving a review. Your feedback helps develop better content. More Jiwenji Creations
Versioning history
major.minor.patch
major indicates a change in at least one of training data, tagging, or LoRA generation methodology. You can use the prompt from any image with the same major version and it will operate under similar conditions.
minor indicates nothing important unless indicated in release notes
patch indicates the number of training epochs, default is 1
Description
Smaller model size
more flexible prompts (more of base URPM model weights preserved)
Further refined training data & tagging
3 epochs x 200 steps, low learning rate
FAQ
Comments (12)
Good evolution of a great mix.
Can you add, if possible, a smaller 2gb pruned fp16 no-baked-vae safetensor version ?
Decent start but it feels restrictive. There doesn't seem to be room to prompt. Modifying or adding to the example prompts and it collapses. Not much of the prompting abilities of URPM or SD1.5 left in this.
@pitreposti792 Please check out the newest version 3.0 and share your thoughts! The prompts are much more flexible.
How come you decided on a full blown model instead of a LORA or a TI?
Making a LoRA or TI for this subject might be possible, but a checkpoint can be further trained. This seems more useful to the community.
@jiwenji would appreciate this in LoRA
@randomseed_co Unfortunately, it does not work, makes no difference if it is used or not, tested the LoRA extract against multiple checkpoints. Don't hold your breath for a LoRA release of this.
nah, i dont think lora is good idea either, most checkpoints are not trained towards NSFW interactions, using lora on top of them will only create distorted results.I've tried them and it is a fact. OP made the right decision
@randomacc There are literally 100's of NSFW interaction loras that work perfectly well.
its not either or, it should always be both. creator is right adding it as a merge add diff ingredient but op is right to ask for lora/lyco too to use some of it on the fly or support some other tools. while a lora might not always give intended result (in this case on none nsfw trained models, but it actually still often does quiet well) it can still help to achieve something or push in a direction or combine with other stuff. imo (depending on the subject) every creator should always try to provide both, including all possible info (training settings, datasets, model base, mix recipes, etc) because that helps evolving community driven things like stable diffusion as a whole.