>>54490 Good lord, AI really is only good for pictures and text. No human being talks or moves like this.

>>54490 Has anyone tried WAN 2.5 for ABDL vids?

Ew gross

>>54547 Their image to video seems very much okay with ABDL stuff. The first frame is from Whisk, this is the prompt I gave to WAN: The woman in pink stares at the screen, her hands exploring her thighs and diaper while she babbles babyish phrases affirming she needs her diapers. It could probably be improved with more specificity, but it seems good to me.

>>54508 And even just still images and text it has trouble with so needs a lot of handholding and editing. It's still mainly only good for memes, shit posting, and easy low quality porn or stock images for anyone with functional pattern recognition. It's insane there's people convinced this is in any way conscious and killing themselves over their chatbot waifus. >>54563 >seems good to me The video may as well be poorly photoshopped gif with how stiff and unnatural the movement looks. The voice is so bad it sounds worse and less natural than just Windows Text to Speech. There's both light from the front of the TV and a screen visible on the back for some reason. The background, foreground, and girl are all out of proportion because the AI has trouble with scale and perspective. How is this "good"? How autistic are you to think this is "good"?

>>54473 What was the prompt for this? It's pretty good.

>>54563 I like it anon, dont listen to that hater

>>54563 how do you get the credits for the WAN site anon? I am not paying for porn

>>54574 You can generate even without credits, it just takes a long-ass time. You also get 10 daily for free. I just use a burner account

>>54573 I mean I realize it's not the best thing ever, but with how censored some of the free models are, the fact that it's okay with touching diapers like that shows it has some potentials

>>54492 I am obsessed. Here are some more fake diaper ads and other clips I made, had to compress. Audio / voice acting is not as good as Veo3, but I think Sora is really good at most everything else. Not perfect of course, and you're not going to get any sexual content out of it unless you're really creative with your prompts, but enjoying this stuff depends on how much of a stickler you are for the AI actually being "realistic" enough for you to enjoy it. For me, I appreciate the progress made in this technology in being able to bring fun ideas to "life". I also think what these two are doing with Sora is cool: https://www.instagram.com/robo_nursery/ https://www.youtube.com/@BlushSocial/videos (shorts and videos) >>54563 Voice is not great lol, but the video does have potential like >>54582 says. WAN at least seems like it can be uncensored and ran locally I believe? I want to try using this next to see what can be produced. Can't get anything else to have the subject touch the diaper at the crotch, believe me I've tried.

>>54573 Eh, doesn't really seem useful to pretend AI generated video and images are better than they actually are either. There's okay stuff out there, but not anything would call good without a lot of editing and work. AI generation makes very basic mistakes because it just has digital training data and no real world reference. This is what annoys me the most about artists and writers who are zealously anti-AI. If an artist or writer is worried about AI stealing their lunch and not seeing it as a tool that still needs a lot of human involvement they're probably not a good writer or artist in the first place. >>54590 And yeah, the voice is the worst part. If this were just a still image and not a video I'd say it's okay for AI, but not good or great.

>>54581 oh man thanks, this is awesome the one without sound I generated with Whisk, it was supposed to say "well look like someone wants to be my baby after all" the other one I generated with WAN 2.5, there are soooooooooo many sites even with similar names for the tools nowadays I am suprised I managed to find you were referring to

Does anyone have any tips on inpainting diapers? I have pictures that I would like to inpaint diapers into, or alternatively to use a person/character's appearance as part of the prompt and create diapered scenes involving them. Preferably local as don't want to share this with google lmao. Have access to decent hardware. Wondering: What tool to use? What models to rely on? Workflow for inpainting?

>>54662 qwen edit probably

Whisk may be done. I made an account a week or two ago and they just asked me to verify my age to use it.

>>54662 I use comfyui and this workflow for flux. It works well to inpaint diapers as long as your original image already has a diaper or panties but you really have to mess with denoise to get it to remove pants. Foocus is also goated for inpainting but it doesn't support flux which most of the good diaper loras are made for. There are forks that supposedly work but i'm too lazy to try them. My image was a wan2.2 T2I that I inpainted with flux and I think it looks great.

>>54721 It's a Google account, just use a fake birthday. Whisk works great.

>>54662 The fastest I've done is by using ComfyUI and Flux Kontext. Works with diaper flux loras, but the diapers won't look nearly as detailed because Kontext, while still flux, is a different model. However, it does a good job at removing bottoms / pants and replacing them with a diaper just using a prompt. It can even add legs or a lower body where it didn't originally exist. (if you make the output image larger than the input) You can do inpainting in regular flux to fix up the diaper and add the detail you want later. Main prompt is: remove her bottoms and make her wear an abenaM4 diaper secured with blue tabs and her legs exposed Workflow and examples below. Modify accordingly.

>>54731 Yeah Whisk is incredible, but they won't let me use it. I have a few accounts and they want me to verify my id to use whisk. Which means sending my ID or a photo which obviously I'm not going to do. I would have set my age to be really old anyway during my account making process anyway. I'm curious how you get to keep using it. I need a way to bypass this forced ID thing.

>>54732 I find that kontext can really mess with image quality and have stopped using it. It's easier to just use another, better, model like wan to qwen and just inpaint with flux.

takes work to go around the limitations of both whisk and WAN but eventually you make something out of it

>>54439 what model did you use for this

Has anyone played with Wan Animate yet? It works pretty good. Although the hugging face workflow seems to speed up some videos.

First attempt at AI generated photos.

>>55072 Chroma

slapped together a lora for chroma, using Miku's dataset. just wish there were more lora's for chroma.

>>55361 Would you mind sharing it here?

Video version of a ai remaster i found of a okayokok comic.

>>55539 > even AI can't redeem OKx4's art

I used to use tensor.art a couple of months ago, but then lost interest in making new images. I tried to come back, but found out that they removed all of their ABDL loras, and can't make a good image for shit. Is there any other site out there like tensor.art with a lora library? most of the sites I've found let you put the prompt and that's it.

Don't end up like them

Anyway to get past Civitai blocking the Diaper prompt?

>>55692 For context: https://inkbunny.net/s/3720704