v3_illustrious have 2 files:
zoda_v3_illu_hires.safetensors (435MB, Hi-Res version , default option)
zoda_v3_illu.safetensors (218MB, Standard version )
Recommend all illustrious user give ANIMA (66MB) a try
Credit to @Choco_Tarte for v3 Beta test.
>> 🎉🎉 Checkout our Zoda Gallery 🎉🎉 <<
Zoda is a LoRA model designed to enhance detail and expression in NSFW generations.
Recommend Setting:
ILLU / PONY / SDXL :
Sampling : Euler A / DPM++ 2M SDE
Schedule: Karras / Exponential
Steps : 20+
Strength : 0.5 - 1.0 (0.5 if distorting in multi-LORA)
Anima :
scheduler: Simple
Steps: 30 - 45
Strength: 1.2 - 1.8 (v1) , 1.0-1.2 (v2)
Tag means work best on tested model base, didn't mean only work on that model.
Illustrious / NoobAI / SDXL are Compatible, cross-apply Lora and models may give out some surprise🎉.
Recommend SDXL ❤️
Welcome to adapt to any SDXL branch, please post it and let me know. ❤️
This is my first model, and while it's not fully refined yet, it serves as a solid foundation for future improvements.
Feedback is always welcome — feel free to drop a comment below. Your input is greatly appreciated!
Description
recommend strength 1.3 - 1.8,
higher strength -> more colourful
FAQ
Comments (4)
great work!
The concept looks great, and it's excellent at bringing out sexy little details, but I find it has an unwanted effect of blearing out micro details in the skin and face. Perhaps this is ok for certain styles, but it goes against the sharp semi-real aesthetic I'm going for. After looking at your concept images, I can see that this bleariness is present in most of the faces shown.
I do not have any such issue with your other ZodaPlus detailer. Is this an intentional side effect of the NSFW detailer?
This is for anima btw. If this was illusion I wouldn't care because I could easily sharpen up the image with aDetailer, but I find that the aDetailer has much less capability with anima...
Thanks for reporting that in Anima released. I usually use aDetailer (yolov8) on Illusion too, but haven't tried it on Anima yet.
I’ve noticed blurriness too. It’s an inherent flaw of this specific detailer because the 1100 x 1600 resolution training dataset already had that soft quality. ZodaPlus is sharper because it used a much cleaner 4K dataset with sharp edges and clear contours. I'll be using a sharper dataset for V2 to resolve this.
Through some bounties I've hosted, I've seen users bypass this with some cool post-processing tricks. For example, running Zoda and Zoda+ together, fine-tuning with FLUX, or doing an img2img pass with a sharpening LoRA.
If you find a better workflow for Anima, please let me know—I’d love to hear your findings! 🥸
@al123sdaw I'm not sure what "sharpening loras" you are referring to.
For now I am running a second sampler pass, with a model that does not include that lora. It's something I do for style anyways and it seems to help out with this. Biggest problem with this is that quite often I prefer to facial structure created by this lora (on lower settings) and doing a second sampler pass can weaken that.