Unlock xxx images using flux, Qwen or Wan 2.2.
ATTENTION
Second version for WAN 2.2 T2V works with the usual models BUT for T2V I recommend some different models and LoRas:
https://huggingface.co/lightx2v/Wan2.2-Lightning/blob/main/Wan2.2-T2V-A14B-4steps-250928-dyno/Wan2.2-T2V-A14B-4steps-250928-dyno-high-lightx2v.safetensors instead of wan 2.2 high
https://huggingface.co/eddy1111111/WAN22.XX_Palingenesis/blob/main/WAN22.XX_Palingenesis_low_t2v.safetensorsinstead of wan 2.2 lowRemoved this recommendationNO speed/low step loras on high noise phase
https://huggingface.co/alibaba-pai/Wan2.2-Fun-Reward-LoRAs/blob/main/Wan2.2-Fun-A14B-InP-low-noise-HPS2.1.safetensors with STR:1.0 on low noise phase
https://huggingface.co/lightx2v/Wan2.2-Lightning/blob/main/Wan2.2-T2V-A14B-4steps-lora-250928/low_noise_model.safetensors with STR:1.0 on low noise phase
???
Profit!
All my videos have workflow embedded, the workflow is just a suggestion. But it works, and works very well.
First version for KREA.
Start using: 20 steps, res_2s, bong_tangent, STR 0.8
After that you can experiment with other arguments. It is not very good and deform hands and feet very badly.
ALL PICTURES ARE PNG WITH WORKFLOW INCLUDED
It is not perfect, sometimes it creates monsters, detached members and strange things, just try another seed or adjust your prompt. Eventually it get it right.
Examples:
Recommended Usage:
Checkpoint: flux1-dev
Strength: 1 (adjust depending on the image)
Sampling Method: DPMPP_2M
Scheduler Type: BETA
CFG Scale: 4
Steps: 35
Feel free to share your creations and feedback so I can improve this model and continue contributing to the creative community!
Description
In this version, I worked to improve the resolution and anatomical details. Also, since some people asked me to separate the realistic and non-realistic styles, version 6 uses only realistic data. This will drastically reduce anime or cartoonish results when they’re not specifically prompted. It’s not perfect, but definitely another step in the right direction.
FAQ
Comments (31)
V6 version，looks like change my lora face
Thank you for reporting. I'll look into it.
seems to have a different sweet spot than the recent versions. more text guidance seems to help (caveat, i have low ram so use turbo alpha lora for most gens). also adheres to more prompt details, imo, so don't 'fix' that, ha
@Narz I'm testing some characters loras from civi. It looks ok on the first tests. @DG9208 do you have any more info that would help me to reproduce your problem? You did not post your lora.
@alcaitiff whatever you're doing keep it up, maybe even offer an online class, most uncensored Loras need your help, hehe. like, sure porn and stuff but it's exactly what most 'modest' models need for any kind of coherent artistic nudes 👍
it is happening also to me, to some Lora that I trained on flux dev. It changes a little bit the face charactheristic.
@jjj1238834s It's hard to know what is happening on your LoRa. But changing the str of yours and mine did not solve it?
I did some test putting less stronger more or less it works. What I notice that for some reason it changes a little bit the shape of eyes or make up. On the v5 this doesn't happens, and I used always strong 1. Thank you for your work.
@jjj1238834s Yeah, I understand. Each model changes something, for now just adjust str a little, on the next I'll try to weaker the faces so it do not compete with any face loras.
First of all, thanks a lot ! Simple question tho, why my generated images do not look like this ? It looks so "AI" so flat, no details, i'm using everything that is said, same scheduler, steps, sampler, everything is like it should, and still it doesnt look "real". even when adding realistic lora, it's still "flat", do you have any text to image workflow to share ? On my workflow i can't change CFG i have "SamplerCustomAdvanced" and not Ksampler, maybe it's the problem ? please help <3
Thank you for your kind words.
About your problem, looks like you are using comfyui so just get a picture of mine and throw in it. All have workflows included.
If you want a very good wf, I recommend my Mystic WF v2. But my sample pictures have a very simple WF. It do it all in one go for 2 loras and show the differences.
Did you try to put one of my v5 vs v6 pics on your comfyui?
@alcaitiff I'm probably retarded but I can't see any WF on his lora page 😅 anyway I found your Mystic WF. It is.... M.A.S.S.I.V.E ! I'll have to test that on Runpod. My AMD RX6800 16gb VRAM won't run this 🤣😅 thank you very much I will test this.
@vinnyvogel70496 It runs on 16GB easy. But download this pic https://civitai.com/images/83957683 drag it to comfyui and drop. It will open a simple WF for tests.
You will need to remove the math part for prompts and input it manually
Would love to discuss some prompting tips with this great Lora. In particular when it comes to action between 2 or more people. Though I don´t fully understand yet, what the model is able to understand. I am having success with these prompt types:
- Ensure the hand is where it is supposed to be:
"with her hands near the man´s crotch, the woman grabs the..." this increases chances, that she grabs at the "right location" and not holding the "thing" somewhere it makes no sense :)
"with her face near the man´s crotch, the woman is sucking..." same thing for oral stuff.
- Using terms like "she grabs" instead of "she is grabbing" seems to have bit more success, but could be placebo...
POV pictures:
"From your ego perspective with your lower body visible, a woman .... A POV photo."
- increasing steps to like 45 and CFG to 4,5-5 can help with more complex stuff to pick up.
Happy to hear about your findings.
any details on the differences between version 5 and 6?
Version 6 is less prone to generating unrealistic images—though not entirely free from them. It has improved anatomical understanding and generally shows better adherence to prompts. On the downside, some users have reported that facial LoRAs might behave differently, but adjusting their strength usually helps. I've also heard that certain checkpoints may produce unexpected results, though I don't have specific details about that.
Appreciate all the work and effort you put in. So I'm not sure what I'm doing wrong but the teeth come out weird in v6. Is there a LoRA I can use to make the teeth look better? Idk how to explain, they appear yellow and almost like shark teeth with large gaps especially the bottom row of teeth. Any ideas?
Please DM some prompts that are giving you trouble
This LoRA is QUITE amazing. I'm 10 generations into using it and it's already solving for a lot of issues I've been having with FLUX up until this point. Seems to work great with Flux1-dev-fp8 and flux_realism_lora.
Thank you @EryxAtys I'm glad you liked
The model is great, but it still affects the character's face a little. Slightly changes the character's face. Can you make sure that the laura character's face doesn't change facial features
Yes I working on this for the next version. People reported that changing the str of lora and xxx solve this problem. Play a little with str and it should be fine.
On v7 I hope this will be solved.
@sm98718 I have a new version and I want to test it with a lora that give you trouble is it possible? Just send me a link so I can see if the issue is solved.
Yeah, people reported this. I'll test face loras on the next version before posting it. For now, we can get good results changing the str of both loras until get the right balance. I'm sorry for the inconvenience.
@alcaitiff Looking forward to the next version. Thank you.
@keyyyyy I have a new version and I want to test it with a lora that give you trouble is it possible? Just send me a link so I can see if the issue is solved.
v6 seems to have a lot more homogenized anatomy between body types, it's harder to get thinner women or non-muscular men, for example.
Noted!
@alcaitiff My early testing of v7 has restored, even surpassed, v5's variety. Still early, but I'm very optimistic. I'll comment again if that changes, for new please accept a hearty appreciation for the improvement!
@ptdurnvcoedqegsdnw Thank you, I'm glad you liked.
Details
Files
MysticXXX-v6.safetensors
Mirrors
Mystic-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
FLUX_X_MAX.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors
MysticXXX-v6.safetensors