V7 may creat body horror. It is not as stable as the v5 version.

V5 is still the best version it doesn’t completely ruin the styles. Why don’t you just expand the dataset for version 5? Versions 6 and 7 ruin the styles and concepts, everything looks like a photo.

Final Conclusion to the point, to not get people bored:

- Very Sampler-Dependent: "res-2s" worked well removing as much body horror as possible (not 100%), some samplers generate horror More than Others.

- Very Strength-Dependent with Custom checkpoints: Maximum strength 0.5 OK, higher than that ? you are gambling, at your own luck !

- Clear winner amongst v5 or v6 or v7 ? NO , each version won or lost in different poses & scenarios !!!

- So, a Personal Taste Choice ?? for me, V6 , it worked at a balance between correct anatomy & tiny skin details (tested on DarkBeastv6 & NSFW v3Flexible Checkpoints) ..

@alcaitiff V6 and V7 report in CivitAI metadata the same "Version title: v6.0"

The CivitAI API gives the same URL for both. https://civitai.com/api/download/models/2834408

how to use this lora? she do ONLY body horror at 4 different zit models. other loras doesnt have this problems. no guide, no any information.

Are there workable solution to use this with character lora? Realtime Lora (etc) not working.

Will you consider an Ltx lora? or is that too naughty these days, maybe just SAy its t2v? <3

I'd like to know how you managed to train a LoRA without altering the original character's face. I've tried cropping out the character's face, but the LoRA still changes their facial features.

Please show some love for z image base new mystic for z image base please

First something positive: I'm not sure how but I do get better variety when using this lora. Not all faces are the same.

Bad thing is, even with really vanilla prompts I get lots of weird body parts. Not just genitals, either. Six or four fingered hands, amputated arms, extra legs, weird mutilated feet, eyes that don't line up or eyeballs pointed in different directions.

I tried to use the workflow but get an error. I can't remember what it is, but it's not just the standard missing nodes. With my own workflow but using the seed, sampler, and prompt from the examples it usually gets pretty close but I might have to do ten generations to get something I don't immediately delete because of some deformity.

Maybe it's that way for everybody, or maybe it's just me. I don't know. As far as NSFW stuff goes, I pretty much never have any luck except for Qwen Rapid AIO. Combining that checkpoint with ZIT works most of the time, but it's slow.

ZIT Mystic v7 might be the newest, but I'd still take ZIB Mystic v1 any day. Any plans to return to ZIB?

I retract my review for the Mystic XXX. This was a mistake on my part, as I tried to use it with non-standard versions of ZIT and those were prone to their own weakness in working with LoRAs and I blamed Mystic XXX. I wholeheartedly beg for pardon from the creator. I feel sorry if my review mislead someone beleiving that this LoRA is bad. What I can say, the reason for this bad review was because every other LoRA (around 6 or 7) I tried worked well but when I tried Mystic my output was botched so I thought this LoRA is the culprit. But with standard ZIT it has been amazing and I am loving it. Thank you alcaitiff and sorry again. I wish you best.

anyone tried this with moody? does it work? does it help?