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
Increased the dataset by 50%. Several new concepts included, better anatomy, better compatibility with other LoRas.
FAQ
Comments (32)
What concepts do you train and recommend for gens? This looks great!
I'll try to create images with prompts to show case it but it will take a little time. I trained threesomes, foursomes, footjob, ahegao, one breast out, lifted clothes, cupping breasts, cowgirl, reverse cowgirl, cumshot, POV, and a lot more. It is not perfect but is improving slowly.
@alcaitiff great, I'm looking forward to seeing the variety in examples. Even if you can include a general guidance in the description for certain concepts. I look at the training captions and try to figure out as best I can, sometimes it's not enough
My best NSFW lora!!!!
Thank you. I'm glad you liked.
Ok cool
One thing I love about this NSFW loRa in particular is it's ability to still generate clothing regardless of what the denoise is set to. I love generating topless/bottomless content, but also like to utilize the details used in these loRas for lewd SFW content aswell. Excellent LoRa, definitely a must have now.
I'm glad you liked it. Thank you for your kind words and for the buzz @AICuriousity22
Great Lora but i find this has a tendency to go for cartoon/anime style when not prompted for. Any way to force it not to render in that style?
I'm glad you liked. To force a style you can use any other lora with the desired style. On this version I put more styles so it know how to draw human parts in different styles. Changing the prompt may help too, for all my examples there is prompt and nodes to help you. And if you find difficult for some specific scenario let me know. I'll help you.
@alcaitiff Thanks, i've had some success with changing the prompts and running a couple of seeds. It's a really great quality lora.
It's frustrating constant cartoons. Pitched this after 20-30 image gens. This could have been prevented or actuated on demand with proper clip training and trigger words.
I'm sorry if you get frustrated @dnmiela772
Use a realistic style lora in conjunction with this one and you will get consistent realistic results. Several ppl asked for a lora capable of creating both realistic and other styles. So it must be trained with both, as some anatomy parts are described different in each style.
In the future I may create a version that do only realistic to be more aligned with your expectations.
OK v6 should be more realistic.
One of my favorite Lora, it is incredible flexible, feels almost better like any pony model. I can´t believe what great pics it generates. Looking forward for updates.
Thanks so much for the kind words and support, @tremolo28 !
I’m still not quite satisfied with it yet, but v5 is currently in training—fingers crossed it turns out better. Training times keep getting longer while the improvements are getting smaller with each version. I feel like I may have hit the limits of this size with v4. For v5, I'm testing a higher dim rank, so hopefully that gives us some room for more noticeable improvements.
V5 is out
@alcaitiff thanks mate, already in use. It is great.
@tremolo28 thanks
One of the best NSFW lora for Flux. Can't wait for V5...
I'm working on version 5 right now, but it's turning into a bit of a monster. Each training session is taking days, and my failures are piling up faster than my laundry.
Still, I'm holding out hope for this latest configuration—maybe it'll finally break the losing streak.
My computer even sent me a message today: "I'm tired, boss."
One day, I'll buy a better GPU. But for now, this little trooper's gotta keep grinding. 💪💻
You can put up some GPU donate campaign?
@AntiLeecher never heard of GPU donations for small lora makers.
@alcaitiff I've seen at least 2 ppl do that. So why not? :) And I think your works are amazing, small but better than many large size lora out there.
V5 is out!
It is the best lora for what it does, I didn't expect much from the file size but somehow this manages to do what most other models 5x in size can not.
I'm glad you liked. Thank you for your kind words.
For some reason, v4 doesn't work, but v3 does.
What do you mean by doesn't work?
look this https://civitai.com/images/80531315
@wolfcatz I think the issue might be how you're prompting—with tags instead of actual descriptive text.
Flux seems to vibe more with stuff like this: https://civitai.com/posts/17899108
This images I generated on Civitai, so it should be easy to replicate and test out.
One without your trigger words, one with it and one adding my lora for style.
Give it a shot
oooh,em...,but ,thanks! good lora!