We're experimenting with AI. Let us know what you think.

[Daphne]

I did an AI-generated movie. I'm curious what you think of it!

[daddancer]

This looks great.

As long as the actresses are real, and AI is used to manipulate/enhance their appearance, I'm all for it. Anything more than that, and it's a "no".

Speaking as someone with a unique kink (noses), I love the potential for "pig nose transformations" like what MissaX did a few years ago, without the need for prosthetics.

[Lucax---]

What model did you use ?

I think you can get better results with newer \ more expensive models (unsure about the NSFW-allowed aspect of it though)...

[witchman]

     Sorry, but I did not like it, nor do I think you should go down this slippery slope.
     Her face did not match the emotions of her voice, her movements were not natural, and her breasts too exaggerated.
     Granted, this might appeal to some, but it did not appeal to me.
     I am not sure AI creations should be used to replace flesh and blood humans. Part of the appeal of your movies is that your female leads are really hot.
     To see them replaced with cybernauts would not be erotic...

[Daphne]

The workflow that we are currently using is:

• The base image (the first keyframe) is generated using Hailou.
• The video is generated using Kling.
• The voices are  generated using Hailou.
• The lipsync is done by Kling.

Hailou's image generation is much better than Kling's for this kind of thing, while I like Kling's video generation better than Hailou's.

The voices are definitely a weak point. Right now, the main reason I'm using Hailou instead of (say) ElevenLabs is that the censorship with ElevenLabs is over the top compared to Hailou (which, weirdly, seems to be completely uncensored at this point).

Every step (with the exception, weirdly, of Hailou's voice generation) is censored pretty strongly. The most frustrating is Hailou's lipsync, which randomly rejects audio based on criteria I don't understand. My guess is that it is doing speech-to-text and rejecting it based on keyword searches in an unsophisticated way.

Some observations on working in this "medium," if you can call it that:

• Porn-friendly generators are really not even in the same league for image quality as the big commercial sites like Hailou and Kling.
• If you do not have an unlimited plan for image generation, you will go bankrupt. I have to go through dozens upon dozens of images before I get the right one for the initial keyframe.
• Kling's prompt adherence is pretty good, if you can stay within its censorship guardrails.
• The voices and lipsync need more work.

If anyone has technology recommendations, by all means, let me know. Remember that everything is censored, and my ability to work around the censorship is a big consideration.

There's no chance that we're going to use this instead of live production, but it would be silly not to at least investigate it. And it was fun putting it together!

[Mesmer7]

No AI can ever get this right. The emotional content of the dialog and vocal inflections are out of sync with the body language.

[foobar]

I think a big thing with AI will be for fetish talent being able to generate a lot of short content/customized videos. Also, they will now have the ability to keep making content where they appear to 20- 30 or 40-something for the rest of their lives.


I am surprised that you are finding Elevenlabs censored. I use it to generate a lot of my personal use video voices, and it will say some really filthy stuff. "cock, suck, master". Kling will absolutely not sync smut, though, but there are obvious tricks that involve some post-production and homonyms.

You can also use LatentSync, which can run in comfyui, but the syncing is somewhat cruder. The upside is that you can run it on a home 4090 setup, or pay comfyonline.app for credits- there is no censorship.

I have been using Artflow to generate base images:

https://app.artflow.ai/my-creations/5c3e10a95912468cacd3fda1632f142c

You can pay them money to train a character LORA and get consistent output. I have had some success training people LORAs on a local stabledifusion setup, but I am not as good as artflow.

[Daphne]

Thank you! How do you find the censorship on Artflow?

[Daphne]

Oh, and I must mention The Worst Thing Ever about Hailou: Disappearing videos. Often, it will generate a great video render, which will then disappear from your account seconds to minutes later with a scolding "this violated community standards" box (ahem, who generated it?). This means you have to watch video generation like a hawk to hit download. They have an appeals process, but that works like 1% of the time.

[grey_shadow]

And I thought Wendy Wild was bad at acting...

Even without the limitations imposed by the built-in censorship, the inhumanity of the homunculus makes it unerotic (at least for me).

[Daphne]

And I thought Wendy Wild was bad at acting...

I have a polite request, which is that we can discuss this without lobbing gratuitous insults at people who are no way involved in topic being discussed. Thanks.

[hishnga]

I am afraid I am on the "no" side of this.

Voice does not feel like it matches the image.  Not a lack of sync but that the female imagined would not have that voice.

It feels a bit like the the poor dub jobs from martial arts movies in the 1970s.

I think it might be interesting to get a live actress to try recording the sound track to the AI generated movie.  That might work better.

Just my $0.02.

[Daphne]

Not a lack of sync but that the female imagined would not have that voice.

Do you mean that a character in that situation wouldn't have that voice, or a character who looks like that wouldn't have that voice?

[David Zeiger]

Interesting. The unrealistic beachballs at the end certainly didn't appeal to me, but as I thought about it, that might be the point? Certainly unrealistic body proportions along those lines are an appeal to many--and this may be a way to integrate some of the more outlandish aspects of our fetish and related fetishes that are otherwise very hard to handle with live actors and a visual effects budget of a small fetish porn studio.

In general, it's still too AI-diction and movement to probably be a viable product right now, but I do see the potential, and who knows what the next round of upgrades will allow.

[lods]

The AI videos that I've actually enjoyed are more short little transformation sequences like https://www.deviantart.com/yamifaves/art/Pushing-the-Boundaries-gif-1210104913

The key art does fairly good white out eyes and the video does a reasonable job at cool transformation effects. No idea how this sort of thing gets around the no naughty images filter most AI have.

It does something that you can't really do in rendered comics easily so might be an interesting thing to do in conjunction with rendered comics? Dunno.

[hishnga]

Character who looks like that wouldn't have that kind of voice.

[boggler]

No AI can ever get this right. The emotional content of the dialog and vocal inflections are out of sync with the body language.

Today, yes. AI will never be worse at a task than it is today.

[Alei]

Ok, just my take.

First, I think we are dealing here with distinct undertone of AI bias, wherein the bias against AI is coloring the evaluation of the end result wherein it doesn't really matter what is being produced, it's going to be bad because it's AI. 

Now, I have certainly seen live acting that is worse than this, and considering where AI was just a handful of years ago, what you have here is astounding.  Compared to what might be achievable with live acting, it likely leaves much to be desired (depending on the actor and the script), and I have some inkling as to how hard it can be to get AI to do anything you actually want it to do the way you want it to do it, so getting anything that makes any sense is an achievement on its own.

I think what you've done here, for you 1st attempt, is really good and I'd encourage you to keep refining what you are doing because, just like anything else, what you are doing will get better with practice. 

An for those detractors who just simply object to efforts like this because it's AI, this technology isn't going anywhere and people in Daphne's position are going to have to deal with, or be buried under this technology as it continues to advance.  Adapt or die people.

[agentsmtih]

I’m definitely more horrified than aroused. The uncanny valley is just too deep

[Tinavamp]

I agree with Alei. I hadn’t commented on the clip before because the transformation portrayed is not one that appeals to me. though my thoughts were, AI could bring benefits to your business, displaying scenes you could not achieve with a live actor. As I am certaint there are many of your clients for who these types of transformations would appeal to.

I can also see the point of some of the other commentators. I fear a world where there are no longer live actors required. I feel the same way about art. I never want to walk into the art supply store and discover they no longer have paint and brushes for sale because no one uses them any more.

You have a business to run, I can imagine it helps put food on the table. There is the saying, innovate or die. I don’t really know you much at all, though you do impress me as kind. I certainly do not want to see you die.

[PlumCherryPeach]

I'm ngl every part of it gave me chills, and not in a good way. Maybe AI will be good enough to make realistic porn one day, but this just made me feel kinda ill.

[Daphne]

The unrealistic beachballs at the end certainly didn't appeal to me, but as I thought about it, that might be the point?

It is kind of the punchline.

[Daphne]

Thank you all for the feedback.

Just to clarify, we are not considering this is a way of stopping live production. This would be to do fun quick little shorts that would be impractical to do with live production.

I agree that the audio needs work. Getting a good performance out of an AI generated voice is still an art (there's a reason that overdub companies that use AI are still able to charge a lot). There are complications in the production pipeline about having a real-life voice performer provide the audio (for one thing, they cost money, and one of the points here is to do things that aren't a big financial hit).

The lip syncing is still really primitive. I'm sure there are better vendors than Hailou for it.

We may continue to do these, but they're really just my personal projects, not anything that is going to be a big part of MCT.

[Lucax---]

Just wanted to add this... I'm not into body swap but I recently found these videos https://www.youtube.com/watch?v=lP6TvJOv6E0 from https://www.youtube.com/@GhostPossession24 . While still a bit in the uncanny valley, they are one of the best AI-gen content I've seen (in term of consistency and video quality). The guy uses Veo 3 and Hailuo I think. For all of us thinking about generating AI content maybe it's a good place to start (?).
Ps: I don't know the guy, would be kind of fun if it was one of us  .

[Jason Hutchinson]

Casting aside any AI vs. Human bias I have, I'd say it has potential, though I did go into the video knowing it was AI assisted, and having read previous comments.  Based on the commentary (and before I really committed to not being swayed), I expected to have a negative experience, but I didn't.  I found the piece entertaining and would likely watch more.