A full bukkake in just 5 seconds.
Trained on 24 81-frames square videos, all of which were structured this way:
initial state (5 frames) => cumshots (4x16 frames) => end state (12 frames).
Just start your prompt with the keyword "sh00tz", and wan will know what to do.
On the high-noise model: works best without any lightx2v lora and cfg at 3.5 (tested with 4/8 steps): semen looks more liquid and natural.
On the low-noise model: add dr34ml4y LoRa (for example, at 0.7 strength) for better realistic outcomes.
Consistency is relatively good most of times, so you should see a coherent progression between cuts.
Wan may draw a weird background floor. You can add "within the same background" in your prompt and/or "background changes" in the negative prompt to mitigate that effect, or just describe the location corresponding to your source image.
Since the training data focused mostly on the face, use a vertical format (like 9:16 or even 1:2) to include more of the woman's body within the frame.
Description
FAQ
Comments (157)
Loving all these heroes posting I2V stuff lately. Thanks qdr1en!
instant personal classic
This is great. But, does it seem to lighten the skin of dark skinned people when cutting?
Yes, despite black women were over-represented in the training videos, it lightens the skin.
please back this up on huggingface, civit removes these kind of loras
I don't think it's necessary. Similar loras like Oral Insertion, Pornmaster Bukkake or Facecream are still up here on civitai.
This LoRA is one of the best for actually retaining high detail and identity while still allowing for a wide variety of outcomes and expressive states. A lot of LoRA that attempt to do what this one does end up being extremely samey from one video to the next, this one pulls it off great! Thanks so much for this one and keep up the good work!
wow, thanks!
A based man of culture👍🏻
He sh00tz, he scores!
Or something.
Thanks for posting!
amazing lora ,sadly civitai is going to nuke it anytime soon ,can you make a cum swapping lora,
what base model are you using to generate your videos?i like that they look too realistic , can you share your workflow?
you think civiat will nuke it? i'm saving it then lool
No, normally it should stay online - I read the TOS carefully before publishing.
I use quantized versions of the base wan checkpoint. The workflow is embedded in the showcase videos, just drag-and-drop them in the ComfyUI canvas to see it.
@qdr1en curious on how CivitAI chooses what LoRAs to delete or not, cuz i kinda just assume all nsfw jumpcut loras are auto-nuked to oblivion
@hsholmeszero It shouldn't undress the character. With any sex lora, panties will forcibly be removed so I guess that's what makes them break the rules.
Thanks personally for that low-res high-noise gen trick, this is a gamechanger for me
Should have named this Bukkake sh0tz.
Please make a version that isn't just rapid fire cuts. We have SVI now, we don't need to have 4-5 cuts in 5 seconds.
Try lowering the high-noise lora strength a bit, and add "hard cut, scene changes," etc. to the negative. It could make less cuts.
@qdr1en I've tried the negative already, but it still likes to do spastic cuts :D
But thanks for the lora, what it does it does pretty well. Just would love a similar one without the frantic edit cuts.
It's so chaotic. Love it.
Yo this lora fucking rules. It's a shame I can't get it to work with nonrealistic but that's tiny complaint. Incredible job on this one
yeah i tried it too haha... Although it kind of automatically makes an IRL cosplayer of that character, which has its own utility...
I haven't even tried with anime pictures, but that's the next thing I'll do.
@qdr1en so after some testing, I can kind of get a "high fidelity 3D" aesthetic that work well with some images (I just posted one with Dva that works well, I think I found the perfect balance between high and low loras, including another Facial lora. 2D anime images will kind of just render in 3D, which is its own nice effect. This lora is a gem!
I made the discovery that if you want interracial just have a black dude present in the original pic and the lora does the rest
@qdr1en You shouldn't mix if you do. It's the main cause of the stupid plastic skin textures of wan LoRAs. Even 'big' lora trainers are doing big mistakes... Please don't, not after such a great start ahah
@GlowingGuardianGirl Oh ok, I didn't know. But I meant, I haven't tried generating a video with this lora and an anime picture as the start image. I did not intend to train loras with anime.
And indeed generating from anime creates some strange 3d/2.5d look.
AMAZING WORK!!
Has anybody had any luck prompting with this that also shows more of the girls breasts?
This is really great.
Works with I2V 384X480p Perfectly: Use with "Cumshot Aesthetics Lora". The following prompt and the result should be fine:
0s-1s:The girl makes a surprised face, opening her mouth.
1.5s-4s: sh00tz.The scene cuts to a penis cumming over her face.
4.5-6s:The scene cuts to the same girl receiving cum from two penises.
6.5-8s: The scene cuts and shows her face covered with sperm and she smiles to the viewer signaling "bye bye" with one hand.
I put 0.5 in both high and low for both loras.
When you say "both loras" do you mean high and low noise versions of both loras? Meaning all 4 loras in total? Thanks!
@frankgallina593 Exactly! 4 loras 0.5 each.
We need more loras like that 🤔
Perfectly cooked one.
Curious how the training went.
- Training resolution for high/low?
- What did the training captions look like. Detailed description of the scenes or did you just give it your trigger keyword?
Thanks, beginner's luck probably lol.
I trained high and low separately, at 256px only, using ai-toolkit (I rented a H100 on runcomfy.com). I left the default settings. The caption was only the keyword and I cached the text embeddings.
But the hard and long part was finding the right moments to make the cuts. As someone else already asked, I shared the first video of the set here https://filebin.net/3pgq6ihiml521g5s. I created it with ComfyUI so it still has its workflow embedded. (I removed the watermark afterwards).
i'm surprised no one has made a yoda one yet
the yoda test is like the "will smith eating spaghetti" of this site
I got you.
It is done.
Top marks. It can be a bit fussy on consistency, as one would expect, but what a cool LoRA. As a pruveyor of cum videos, I thank you for this. I'll probably use this a lot. Benefits from being really explicit with prompting.
can you make a cum swap lora https://filebin.net/zm8soy89lei1y1z3
Not really my type, sorry. But I have other ideas on my list.
@qdr1en looking forward to the next one!
I'm pretty obsessed with this Lora. all-timer for me. I agree with some other people that it would be nice to have a version with less cuts, to allow for more time before and more time after. I know you said you have other ideas too, so fair play. can't wait to see what you cook up, cause holy shit.
Haha, glad to read that.
Well to get less cuts, maybe you can try adding other loras on the high-noise model (give dreamlay a shot), or reduce this lora's strength (I tried once at 0.5, got 2 cuts only but she stayed alone). If you find a way let me know!
Dear lord.
EDIT: Okay, as expected, it's amazing. I wonder if there's a way to extend the establishing shot.. maybe with an SVI workflow? But if it could happen in the Lora, that'd be incredible- SVI still gives me lots of grief!
I'm doing post SVI extends, then more sh00tz
Feel free to use my workflow (just download one of my latest posts that has "workflow embedded" in the prompts. It doesn't use SVI but automatically stitches segments using different prompts and lora, you can leave it as is or modify each segment to your liking. I'll publish an official workflow after some cleaning and polishing
@Bonz_ai You're a fucking wizard. I'll check it out when I can. I might experiment with SVI, but I used a workflow that gave me major ghosting issues (after a 30m+ render) and it soured me to it a bit.
@artificial_infatuation I haven't tried SVI yet, seems like a headache. I have 16gb vram and can run my workflow in 10 to 12 minutes, which makes a 10 second video, which feels pretty good I think
@Bonz_ai 7900 XTX and have made 30sec vid so far, it works well imo
@benjareynolds690 damn I wish... Btw I just noticed an important mistake in my workflow, for the first generation (the bukkake part), the lightning high lora is enabled accidentally, cause I put it in a group. Just straight up delete that node (just for the high lora), the outputs will be much better!
Can you make a Wan 2.2 5B lora? 81 frames on 14b is rough on 16gb vram
That's not a valid reason. I run it with 12GB VRAM and 32GB RAM.
Use quantized models to solve that. I use it perfectly with 8Gb Vram and 48Gb Ram with quantized Q4_K_M models
For statistics: I have 16vram and 64ram, so I can use up to 40gb (like qwen edit 2511 bf16) full models (with even 32ram it was difficult, but still possible), also it must be MoP model with offloading unused weights to ram (qwen and wan2.2 do that). So, I believe the point is to get as much ram as possible for working with wan, and optimized workflow (wrapper based with blockswap 40 or default comfy nodes - I heard they implemented blockswap directly in code).
Can anyone share their workflow for this? I'm struggling to get results as good as the examples - great work btw!
feel free to use any of my recent videos that has "workflow embedded" in the prompt. Just adjust the prompts according to your input image
@Bonz_ai amazing, thank you. I didn't realise they had workflow embedded, that's a big help
i use https://civitai.com/models/2005198/lazy-wan-22-i2v-workflow?modelVersionId=2379424 and it was ez to setup and the face enhancer works magic
The model doesn't really do jumpcuts/cuts. It just does a few cumshots from one guy on the right and another on the left in one shot. How would I make it so it's more reminiscent of the example videos?
I'm away from my setup right now, but try experimenting with the shift and experimenting with how many steps you're doing in and out of lightx2v. That can make the cuts more dramatic.
@faloje6199513 I figured it out, I use WAN2.2 Remix GGUF and I just had to:
Steps: 5
CFG: 1.0
Euler - Simple
Put this into Positive Prompt first: sh00tz, quick cut scenes
I don't deal with any shift or anything, it's a super simple workflow that I can probably share here soon on my profile.
This is the greatest Wan Lora to drop in recent times! Thank you!
does anyone who's savvy with comfyui know if there's a node that can analyze a video and detect a frame where the person has their eyes open? ^^'
Would really help me with my workflow
Just one node? Most likely there isn't. You could try to input video frames into yolo or some other detecting node with eyes detector model and somehow pick frames with high probabilities, most likely with a vibecoded node.
@zanebe Thanks for the answer! I sorta managed to solve my problem with just prompting since there's no easy way to do this
Incredible! Simply incredible!
very nice one ! well done mon ami !
Suggestion for next lora : same stuff but with 2 girls ?
Thanks.
Hum.. It is not different enough from this one. I'd need something different, publishable on civitAI, and that has not already been done.
Tried with two girls but it ends up merging their faces
What happened to the reverse one?
nuked :D. never mind
@qdr1en What a shame, don't give up please, your work is amazing
@jakoc75648 Thank you!
I'll try something easier next time, and see how it goes.
@qdr1en bummer it was good. Do whatever you feel inspired to make and try anyways.
@qdr1en now i'm curious... What was the reverse one?
@Bonz_ai It was a reverse deepthroat lora I released yesterday, but it was removed.
Can you please share at least the prompt for the nuked one here for the people who could download but not immediately test it?
@Lora_Addict It's on huggingface now.
@qdr1en link / keyword to search?
This lora is F*CKING BRILLIANT! Easily the most genius one I've seen so far.
My question is would it be possible to train a lora that transfers a concept to other "acts," such, for example starting with a face, and going to doggystyle, missionary, etc? The possibilities could be truly epic, but I don't understand how your training works so maybe this is outside of the realm of possibility. Thoughts?
Lol you arrive a few months too late my friend, all those loras were removed from the site at the end of the last year.
Thanks for your review.
@qdr1en Damn! Well can they be found elsewhere?
@frankgallina593 search up "blink" on civarchive, there's a bunch
@Bonz_ai Thanks!
@frankgallina593 RIP your sleep
@qdr1en why were those removed? did they work?
Incredible LoRa. Making one like this for LTX2.3 that actually works well would be fantastic. It's much faster than Wan.
It won't be from me, I'll stick to wan for now.
Can you make one of these that's just for boobs instead of the face?
You're a wizard? 'Cause you've got me all enchanted. I can't stop generating stuff with this LoRA it's so damn awesome. Top tier work. Thanks a ton.
Thanks for the tip, man.
The recipe was a mix of inspiration, some work and a dose of luck 🍲.
@qdr1en 100%, you nailed it
any chance you can also post the WF for one of your samples?
@ffrulez All the showcase videos still have the workflow embedded. Drag-and-drop them inside the ComfyUI canvas. I used a node, "terminate", which isn't available, you can simply delete it.
WOW! este lora es oro puro! gracias el mejor lora de este tipo!
great
hola alguien me puede enviar un workflow para usarlo? el que tengo no me deja usar el lora, y si añado un nuevo nodo para usarlo el video sale borroso...
Es fácil, yo te lo envío.
@Pantyhose_master same, it is blurry
@Pantyhose_master Me lo podrias enviar por favor? Tambien me salen los videos borrosos.
@monxo2000 dame tu mail, o algo por donde mandarlo! saludos
Puedo pedirles asistencia en el mismo tema? Gracias de antemano!!
God Tier work on this.
One slight problem, the penises are all the same, and no lora stack will change them in a meaningful way. (Not in size, shape, color, or "details")
Hi. Indeed, it won't include black penises as there were not present in the training data. But I tried adding "The penises have different skin colors and sizes: they can be light-skinned or dark-skinned, they can be big or small.", and I got different shapes and lengths.
I tried 10 images. They all came out as a video with a little movement. None of them did what your videos show. i dont get it.
Did you add the trigger word ?
Did you use the base wan models ?
Did you run the first steps without the lightx2v loras ?
@qdr1en Should it not work here on civitai? I used the wan base model with the trigger word here with the civitai generator.
@TommyLedoux I don't know, I never used it.
@qdr1en Do you have a comfyui workflow that you could share? I tried it with the comfyui wan 2.2 i2v template but the results are extremely blurry. Would be amazing!
@TommyLedoux There is a workflow embedded in every showcase video. Just drop them into the ComfyUI canvas. There is a node called "terminate" that will raise an error, you can erase it.
@qdr1en Oh man, I tried it and it didn't work. Now, after you comment i tried it again and it did work! Thanks!
Fantastic work. I wish it was a little bit better with the consistency of faces from shot to shot, but still brilliant nonetheless.
Thanks, @qdr1en can you please add a text file in folder on gumroad with example prompt..
You only need the keyword, really. But I will post more examples on civitAI
@qdr1en I was thinking the other models, g4ngfr0nt etc. Nice if you can include some examples in folder, text, etc
This doesn't work with on site generator. Creates blurry images and doesn't generate genitalia correctly.
bro. you gotta get your own rig! don't complain :)
Anyone know if this is on tensor art or tensor hub yet?
Hi. I don't intend to upload the LoRA there.
hey how are you using the patch sage attention kj?
Its not working for me on windows
whats the prompt beside sh00tz? whats the high steps and low steps. and high noise and low noise on sampler and model strength and clip strength for high and low , cfg etc. i want it to work so bad. i type in shoots and she looks at camera and just eats some random fruit or somthing :( no camera snaps or cum, nothing :((( im using wan nsfw remix iv14b fp8e4m on high and low model with the shootz on highand low lora
There is a workflow embedded in the showcase videos, try it.
And use a base wan checkpoint, not a remix or merge.
Help! I got this working but... that last part where it interpolates, I have to disable cause it throws an oom error, I also had to disable the sage nodes cause I don't have sage nor do I want it. Anyways, the video generated completely but it's aliasing bad and doesn't matter how I do the image resolution all finished videos have a sorta doubling kinda didn't denoise properly look. Any idea what would fix this? I've tried so many variations with the ksampler and they all made it worse. Disabled the light2x like you said added the steps, it just got worse. It's like so close to being there it just looks like sh*t. Keep in mind I have no trouble with generating videos other wise. This Lora I've been searching for awhile and it's so close just no cigar kinda thing. Any help would be appreciated.
Hi. Are you sure both the width and height are multiples of 16, at both stages? (look at the bottom of the "Resize Image V2" nodes when they run). That's the most probable error from the showcase vids' workflow.
For the interpolation part, I'd recommend you unpack the subgraph and keep only the "GIMM-VFI" nodes (if they are red, you have to install those custom nodes first).
Use the base wan checkpoints (either fp16, fp8 or Quantized), not a custom one.
And for the settings it should be all good with 4+4 steps, euler/beta, shift 7.
Let me know if/how you manage to fix this.
@qdr1en Thank you for the reply. After some research I found out it was the type of lightx2 Lora I was using, which was the 1022. I was reading that some people were finding with certain Loras they had to use a different lightx2 which is the MoE version. Once I put that in place it fixed everything, no more aliasing or shadowing the videos look great. Again thank you for your time and attention and all that you have done. Much appreciated.
one other question while your here. You know how you said about adding the Dr34ml4y lora for better detail? How were you triggering that lora? It has several different triggers and was curious to which one applies best to this. Thanks
@burn1one1one812 It has several different trigger words but I don't use them as I don't need to. For other usages, better see directly on the LoRa's page I think.
This Lora is amazing. I'm getting amazing likeness stability. One additional thing I've found this Lora can do (entirely by accident) is that you can basically use it to create a fast cut montage of bukkake cumshots of almost anything. What I did was accidently include the Lora in a workflow for something entirely different. It followed my actual prompt but now my ass shaking twerking video had a cumshot progression that looked perfect. Occasionally it will try to force a face into the shot but I've had amazing results with foot worship workflows and ass shaking as well. You don't even have to use the keyword trigger. It's my go to way of getting massive cumshots to cover up whatever I want.
Great lora! works flawlessly. Any tips on retaining facial integrity? face seem to degrade or not resemble original image over time.
one tip i have is cropping the starting image so the characters face takes up more space on the image, face adherence is very high when the details of the face are unambiguous.
Admire u guys. Im just starting on this. Can you help me with a workflow? Im very frustrated...
Txs in advance
Ltx 2.3 please
Absolutely insanely good. Thank you for the training notes, that's very useful to know how it was done.
amazing work. hope to see more like it
Looking for help with the following setting. sh00tz high and sh00tz low and cfg?
1, 1, and 1 :)
any tips on reducing the appearance of eyeshadow, for a more natural look?
maybe try the T2V Makeup Slider ? https://civitai.com/models/2259797/makeup-slider
Hello all! Looking for a quick tip. I'm getting mutant dicks, extra limbs and even horror faces lol. Not sure why. I'm using WanGP v11.13 by DeepBeepMeep. Default settings and HN and LN LORAS Multipliers 1;0 0;1. Thanks for any help.
I'm not the uploader here but I've been having similar issues where it's fine for several videos then they start like ripping their c*cks off and using them like paint brushes or sometimes the c*cks come out of their mouths. The only thing that makes sense is that it would be the seed changing cause that's the only thing different but.... I don't think this lora will let you lock the seed in place I think it might be vital to how it works. One thing I did take notice too as an after thought, that has helped lessen it...is (and I was doing this for awhile without realizing it) is to make sure your taking into account that if your staking loras that the lightx2 4 step lora counts as a lora aswell so hense you might need to adjust your chain of loras to a lower strength because they have to be read through the lightx2 first.
@burn1one1one812 Thanks I'll do some testing.
It should work well with base wan models + 1030 HN and 1022 LN loras, and this lora at strength 1.0. Did you use those ?
@qdr1en Hi, thank you for your reply. Could you please clarify what are the 1030 HN and 1022 LN loras I should add?
Ok you mean the Lightning Lora? I'll give it a try and report. Thanks.
@qdr1en Thanks ! Works pretty good now.
The download doesnt work no more
It keeps adding weird random men's faces with cum shooting out of their mouths... or will have empty hands without a penis and cum dripping from the hand
try the workflow from the showcased videos
sad that the workflow is a bit spagetti so its hard to maintain what where it is .
anyways. does somebody know how to repair or enhance the males in the background? i got sometimes mutants with 2 testicles or none or moving without hands etc.
Are you trolling? The workflow in the showcase videos is clean.
Details
Files
sh00tz_HN_75.safetensors
Mirrors
1-sh00tz_HN.safetensors
sh00tz_HN_75_KEYWORD_sh00tz.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
high.safetensors
sh00tz_HN_75.safetensors
sh00tz_HIGH.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75_copy_test.safetensors
sh00tz_HN_75_global_dedup_test.safetensors
sh00tz_HN_75_no_source.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
shoot-high.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
Wan22_Qdr1en_sh00tz_series_of_cumshots_HIGH.safetensors
sh00tz_HN_75.safetensors
shoots.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
wan2.2-i2v-high-bukkake.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors
Sh00tz series of cumshots I2V HIGH - Wan2.2 - sh00tz.safetensors
Sh00tz series of cumshots I2V HIGH - Wan2.2 - sh00tz.safetensors
sh00tz_HN_75.safetensors
sh00tz_HIGH.safetensors
sh00tz_HN_75.safetensors
sh00tz_HN_75.safetensors