@MakeThemComeAliveAIArt All good man, keep the buzz. I like supporting creators. NF4 does indeed support LoRAs in Forge now as of a couple days ago. You might just put a note in the LoRA description that this is Comfy only so people know. Maybe it's time for me to make the comfy leap anyway! :)

Works on forge for me np

@Kyra31 can i ask what checkpoint you're using?

@MakeThemComeAliveAIArt I got it somehow, idk i cleared everything out and restarted. Might be a good idea to delete this comment if you want so it doesn't give people the idea that it doesn't work.

@MakeThemComeAliveAIArt I can't create a new comment but wanted to thank you for the lora.
Little tip: You can add

"cum is squirts of dripping translucent whitish jelly"

at the end of the prompt if the cum doesn't appear enough. It already worked without the lora but with it it works wonders!

@TipsyThrowaway14 What do you mean by cleared everything out and restarted? care to elaborate? i have the same issues as you and using bnb-nf4

@chrisss1 this is what i did. load the latest forge with dev fp8 using --ckpt or have it loaded when you shut it down. describe the substance as @Error666 describes, and describe where it is. don't load any other loras and use 1.1 or 1.2 strength. worked for me.

@TipsyThrowaway14 The problem is, i can't use the standard flux dev fp8. I have 2060 and 16gb ram that's why im using bnb nf4 v2. I still tried whatever you say but still the same.. can it be work in bnb nf4 v2 too?

@chrisss1 i never got it to work with bnb-nf4

@TipsyThrowaway14 I want to clarify that yes, this does make some jelly like substance appear, but it's NOT a good quality, it's very fake looking, and the image itself is also very "AI". While if you use this lora on comfy, it gives the exact same results as the example images, which are on a completely different level. I do not consider that workaround as a workaround, since yes, technically it's adding liquid, but it's not following instructions and giving very subpar results

@MakeThemComeAliveAIArt I've tried but nope, it only adds a few drops, nothing close to your example, same size, seed, sampling and schedule

@MakeThemComeAliveAIArt I installed and tried it on comfy and it works perfectly, so it's clearly some forge implementation issue, the lora itself works very well in comfy, I wonder if it's a setting somewhere or just a broken implementation

Does work for me. in standard flux dev. Maybe try Afrodite Flux model

@MakeThemComeAliveAIArt Awesome, how are you planning on improving it in the next version out of interest? If you can find pictures of the same people with and without cum on their face that might help to learn the concept better?

@MakeThemComeAliveAIArt I have not tested it, in my head it’s similar to hot Dreambooth works using regularisation images to extract just the concept you want into the token. I might do some experiments with it to see if it works.

I just did : https://civitai.com/models/858262/cum-on-face-flux-non-face-altering

I tried but my wife slapped me

Use a dedistilled version of flux and/or prompt "explicit image/photograph of>>>" then end your prompt with "NSFW" if you're not using dedistilled versions. You could even use something like this "Explicit adult themed pornographic image of>>>>" and this will usually suffice. Also, if none of that works, run it on your local machine because if you're on a hosted site, many times they bake in a censor onto whatever model you're using, and if that's the case, its hard to bypass. Also, if your machine sucks you can run on a pod in runpod. The L4 gpu, for example, is 24g, and it costs about 45 cents an hour to run for the gpu plus storage. They have comfyui with flux templates on runpod already. Aitrepreneur and Valyrian both host templates on there and both work well. Valyrian is a "one click" template and if you opt for persistent storage (you rent drive space for about seven dollars per month per 100gigs) once you used the valyrian "comfyui flux dev one click" template, and its installed on your storage, when you delete that pod (as you would do to prevent being charged for it if not using it) then when you start a new pod, you can use valyrian's "comfyui without flux dev one click" for a much more rapid install onto the pod and you can be up and running in minutes. It may take a day or two of looking around for the proper models, proper nodes, proper loras, etc that you're going to want to load onto your rented drive, but once its set up, its good. My current drive on runpod is 250gigs and it costs me 17.50 a month. It has two flux dedistilled models installed, tons of loras, wan 2.1 models, and tons of nodes. So when I load the "comfyui without flux dev" from valyrian, I'm usually rendering images and videos on the pod of my choosing within minutes, and you have the option of what type gpu you want. Since comfy only supports one gpu, you can use previous gen nvidias, current gen, or a couple of amd options, and you rent them on demand for as little as .28 cents an hour to as much as 3.99 an hour. Keep in mind though, that sounds like "a lot" of money, but you'd have to use the service for 3 hours a day for 3 years straight at the rate of 3.99/hr to spend the equivalent price of one upfront cost of buying a single h200sxm nvidia card.... so its reasonable. The RTX4090 and the RTX 6000 ADA are both under a dollar. I think the 4090 is .69 an hour and the rtx 6000 ada is .77. I don't have a graphics card on any of my computers because they are all music computers and they're all over my studio. So I rent my gpus on demand. BEST PART is, if I'm using flux, I rent the L4, for like .45/hr, or the 4090, for .69, or the rtx 6000ada for .77 but if I want to do a bunch of video renders I do one of the bigger cards because they whip out WAN 2.1 videos (ten second videos) in a matter of a few minutes. So for 3.99/hr, once you've got your prompts worked out (on the 4090 for much less money) you start a pod with a badass gpu and can render as many as 30 videos in an hour. Very cheap, all things considered, because if you purchase a kling membership, it ends up costing almost 100/month just for a handful of videos every month. For 4 dollars, I can render almost as many in one hour as i can in a month on kling. Runpod's not the only gpu server either, its just the one I use.