Lora for Connaisseur only.

@eoa42 What do you mean ? 😭

@K3NK exactly. @xuadamux373 Set your High model to cfg 2.5 or 3, sample the steps without the lightning Lora. High takes almost no time at all. Put "Slow motion" in the negative prompt.

Leave the low models at cfg 1 with lightning Lora enabled.

Some people even use 3 samplers, one with cfg 2.5 to 3, one with cfg 1 for high, and then the low.

I had not a single slow motion in like 50 generations. You doing something wrong mate! And i use speed Loras on high AND low.

@K3NK Yea, figured out increasing the strength of the lightning Lora on the high pass solved it for the most part, strange

@rustybutler Thanks! I figured out increasing the strength on high pass to 3 did the trick but I’ll also give what you said a shot as well

@K3NK It was not against you, although you could release the other model if one reached its goal.. i mean, if you have trouble handling a zip file, you might should stay away from AI :D

Anyway, thanks for the Lora, can't wait to test it!

gift cards still not purchasable. So you have to start a crypto wallet, learn crypto, just to get a lora.

Might as well become a crypto trader

@whitespider9999 yeah, i would immediately buy buzz and push this to release but i will not go through all this Coinbase wallet creating and KYC process just for that :D

@marqs89 same here. it's the goddamn crypto thing that's an obstacle to me, its comparable to the work of starting a small business. Not even joking. IF you want to do it securely and have a long term wallet.

I'm not even against the creators. I think they deserve buzz for all these works of art. I don't think them wanting money so they can keep making stuff for us is bad. It's something i support. And want to support

But why not release both when one reaches the target? You don't have the ability to take it out of early access? Or just reupload as public?

@psspsspsspssspss i dont know the actual reason, one reason could be that this inspires people to pay, since i believe you could use the high lora to trail things out, without really having the whole output quality. (many loras release with just a high or low lora)

@K3NK just upload a copy to this listing and not lose anything? I don't understand why you can't do that. people always upload different version or collections of models to a listing. Are you just holding out for more buzz?

low is released! Thanks to whoever made that possible <3

@K3NK But is it a typo in the training data? Or just the instructions?

yes of course just prompt and see?? if you need to ask question like that before generating i think you might not have strong enough of a machine to do generative ai.... invest money in 50 series its fucking amazing. i make DT vids in 110sec on my 5070ti i can only imagine on rtx 5090

@Ayylmao1773 yes you are right, but what's the point of the community if we don't ask questions here for others also to see:)

yea the facial change part I am starting to notice, its insane lol

I may or may not have a queue of 13 various image to vid gens running... 😐

@whatsthisaithing that's rookie numbers! :p

@whatsthisaithing noice lol

100% agree, this LoRA is amazing. @K3NK: any tips/tuto on how to produce LoRA without facial distortions?

@K3NK yeah this one is kinda of scary good. nobody posting a selfie is safe on this one lmao

Spit/Saliva can be added by a simple prompt

@K3NK Thank you!

@K3NK Thanks, works better!

great advice, what strength have you found works best for this one with penis play?

Thanks, this is great!

@xuadamux373 I ended up settling on 1.5 strength for Penis Play/0.5 for Ultimate Deepthroat on the high pass, then 1.0 strength on penis play/0.85 strength on Ultimate Deepthroat on the low pass, but I didn't test that many combinations. The thought is that the high pass controls more of the overall motion while the low pass controls the final look of things, so I want UD to have more power on the low pass.

@sharp3 Thanks I'll give this a shot

I've posted a few workflows with face restoration nodes. You can use a reActor node and use your input as the source image, or use a reference image. Or saved face models if you have them. If there is occlusion it can get tricky, you can either automate the masking with a face parser or generate two outputs, one clean and one restored, then mask manually in post. Sometimes it's a breeze, sometimes it's a massive pain in the ass. But it's doable, and as with everything else you get out of it what you put into it.

@Ponder_Stibbons Thanks, but I was referring specifically to Lora training. Just trying to make mine better. :)

@K3NK Thank you. This worked perfectly. And removed the random blur.

@K3NK is the high and low all weight 1? and the lightx2v all weight 1?

@K3NK It occurs for me more when the subject's face is a bit further in the image. If it's a face closeup, the lora manages to keep the facial features intact

I've noticed reducing the LORA weight has helped with this somewhat. I've used 0.7 on both low and high and it retains some of the facial consistency better across non-euro phenotypes.

@K3NK yes, I'm using speed Loras (loaded as first after Wan I2V H/L). Strength 1/1, I tried 2/2, but that didn't really work. I tried turning off other loras, changing loading order, changing sampler steps, but it always look very... gentle

@evk3 Maybe worth trying a different workflow, try experimenting with dragging and dropping some of the videos that look good here

@K3NK @Light2020462 I think I got it! A significant progress, at least. I replaced default Wan2.2 light lora with lightning versions and added Wan2.1 light loras with 3.0/1.5 (H/L) strengths. Thank you both.

load both the high and low noise models. You're likely only loading in the high noise.

Appreciate your replies.
I have the Lora on high and low, both at strength 1. (Model Q4 and Lightx Lora) But they still have shitty penises that look like sausages :). I'll play around with the shift and weights.

Tried around, nothing works, add dicks-epoch_100... nope, still sausages. Strange, because if I use the oral-insertion lora instead, everything is fine

LOL. I'm such a twat... did not use the low and high versions in the right places. Now it works

@citivai846 in which node u put them?

I second this. I held off for a while because the thumbnail didn't' seem appealing.