2 Girls 2-Handed Twisting Handjob Frontal MInimax H3 V1
Prompt example:
"The naked, gorgeous girls with perky, naked boobs in Picture1 and Picture2 giving a handjob twisting their hands around a well-formed cock"
Used STR=1
Caveat: The LoRa results in very similar results, more than I expected looking at the training data which looked quite varied. When I tried simple T2V without a description of the girls, I got four clips of almost the same girls, which surpised me. It still works, but you should use either
T2V: Describe the girls
I2V/FL2V/Ref : Gives the most predictable girls. Images for the each girl and for the cock gives the best results.
Another surprise came when I tried to combine it with my previous "Blowjob Sandwich" LoRa and gave the prompt for it, For some reason, the handjob LoRa does works well without need the sandwich lora, just by adding the prompt:
"and performing a blowjob sandwich, a cooperative fellation by kissing, licking and sucking opposite sides of a cock and sliding it forwards and backwards between their mouths."
No extra LoRa needed for those tongues to do a good job.
@BonerSoup just helped me find that lowering the STR<1 with image references will get both better fidelity to the original girls, more varied expressions and a member size that is more to your liking. I just tried with STR=0.5 to excellent results.
Description
FAQ
Comments (31)
Vey nice work! now we need double blowjobe :P
I made a first attempt with Blowjob Sandwich H3 Minimax V1 - BJSandwich H3Minimax v1.0 | Hailuo H3 by MiniMax LoRA | Civitai
I noticed I could get a bit on the way just using this prompt: "Use ref_image_0 Picture1 for identity of the girls and cock.
The naked, gorgeous girls in Picture1 performing a blowjob sandwich, a cooperative fellatio by kissing, licking and sucking opposite sides of a cock and sliding it forwards and backwards between their mouths." I think there may be many ways of doing double blowjobs, and you are welcome to suggest different ways to inspire me to make different new loras.
Only downside with this lora is sound.. are cracking etc but rest are great :)
@Donkeyworld Yeah, H3 sound is a toss-up. You can of course, prompt to get good sound, but I only wanted to showcase the central theme.
I remember double blowjobs being so easy in WAN22 . I just had to present a cock in front of two girls, and they automatically started licking it. Sucking was harder. I don't think H3 does this, but I am right now giving it a test run.
thanks for lora H3
Yes, baby, we need more of that, but so far, no excellent LoRA specifically designed for licking the glans has been released. Even with highly detailed prompts describing the tongue rotating against the glans—or licking the base of the penis—the results are poor. If you can pull this off, you’re a true genius!
I may give it a go.
3 LoRas then:
1. Licking glans.
2. Tongue rotating around the glans.
3. Licking base of penis.
I usually go for 2 girls, but maybe I'll try with one.
Full disclosure: I am waiting for whatever base model comes next that will not destroy all genitals at every opportunity.
@CyberCream That’s really great—I look forward to seeing your work!
@Silicon_Mirage @Silicon_Mirage
After a bit of testing. 1. Licking glans works without any loras. It's not great, but it works. 2. Tongue rotating around the glans is hard. No success in initial tests. I have not tested 3. yet, but we will see whether I do anything or wait for the next model. I suppose it depends on how motivated I am, how hard I will try.
Very good! works well with reference characters and a seperate reference penis as you stated in your instructions.
I am happy for your positive input.
2 girls 1 twenty inch weiner
I don't know why. The training data included large ones, but not that long.
average size
Any chance to update the lora with regular penis size?
Interesting thought. I shall make a test, but I suspect the size strongly depends on the input image.
@CyberCream plenty of small peepees out there
I ran into an unexpected problem. My image generator keeps giving me above average size cocks when I describe it as two girls twisting their hands around it. If you provide me with one, I shall test. Or you could test it yourself, of course, post it here, and link the result in this thread.
@BonerSoup To train, I need two girls twisting their hands around a regular-sized. I have not seen any. So I generate; the problem with generation is that what I generate also gives me above-average.
@CyberCream You mean like this?
https://civitai.red/images/145925619
you can generate plenty with the new MOSS-Image model, it's very good. that image has the workflow in it, drop in comfyui.
@BonerSoup Thanks for the image, which I am testing now, and for the MOSS workflow.
Thanks to you image, I could test. I made two tests, one with STR=1, and one with STR=0.5. I have posteds the STR=0.5 here, and I am happy to report that the current LoRa works well, but one has to use a STR <1. This is a good result, since STR<1, which I had not tried before, keeps a much better fidelity to the girls in the image. Image post by CyberCream | Civitai
@CyberCreamYO! that looks really good! also their expressions are different from each other, which was a problem in the lora, they copied each other's head movements and facial expressions.
@BonerSoup I want to use MOSS now. Where do I find MOSS-Image\MOSS_DMD_Turbo_lora_latest.safetensors ?
@CyberCream https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/tree/main
it's the "lora_latest.safetensors "
I changed the name.
@BonerSoup THanks again. Almost there. the clip thing gives me "Some input values are not available for this node.".
@CyberCream update comfyui to nightly version to enable the "moss_image" dtype option in the clip loader