- This topic has 31 replies, 6 voices, and was last updated 2 weeks, 6 days ago by
.
-
Posts
-
-
I’ve been trying out (in a very limited way given my lack of generation credits) converting photos to videos. I’ve only tried it with photos I own so far but the results have been impressive.
I’m wondering if members would like to see themselves naked in situations they could not or would not be able to experience in real life?
It would involve sending me a good-quality, well-lit nude frontal photo in landscape format. I could then post one or two variations
The results are NOT perfect, but the over effect can be stunning. These would have to be in the age-verified Patron-only area, as there would be nudity and a small cost involved.
-
Gary very kindly volunteered a suitable photo, so here’s a quick example of the sort of thing that can be done. And, before you ask, we can also work with full frontal nudity.
This is a first attempt and very low res as I’m trying to conserve credits. There are glitches but I’m mostly pleased with the result.
-
@ed I’m quite happy to try similar with full frontal nudity. Just let me know the requirements
-
Same spec really.
Good lighting, sharp focus, highest resolution you can and, for this one, no clothes.
Any ideas what sort of people you would like to join you in the video? I can’t specify actual people, obviously, but general age, build, sex, clothing etc.
-
No, go for it, surprise me !
I’ll try and get something to you ASAP (the weather forecast after Friday is ⛈️).
-
Okay – that’s brave!
No rush. I’m unlikely to have opportunity to do anything for the next week or so anyway. Also, the trial subscription I took out has now expired and my remaining credits went with it 🙁
-
-
-
I’ll have to see how things go, and how much interest there is. I’d love to experiment more with the whole area but my hope of getting free credits through a small flow of referrals, hasn’t panned out. It seems no-one else wants to give Pollo a try.
Maybe after August, I’ll (hopefully) be able to afford a proper subscription so I get an amount of credits to work with each month. In the meantime, I’ll create some more examples.
-
-
Not quite “AI generated” and they’re not naked either … but the idea that robots could be programmed to play football intrigued me. Some of the results are even funnier than I expected.
https://www.bbc.co.uk/news/videos/c5ylkyrkjnzo
-
I have used AI to make pictures but so far I haven’t actually tried videos but that will be my next new distraction I suppose!
-
I thought there would be more interest in this. It would have been interesting to explore.
-
- 49
- Experienced Poster
It looks interesting as a demonstration of the technology, but (due only to my bad overthinking habit!!!) I found it quite hard to watch.
- I couldn’t see where Gary had consented to it (though obviously if he’d objected it would have been removed.)
- I struggled to think of a story that would include this episode. I settled on the guy having two friends who have entered an event. The friend in red has done well and is being congratulated by her two friends (the guy and the one in green). Looking at it that way, are we thinking unkindly of the friendships, as if we’re being encouraged to see them flirting and therefore doubt their sincerity in the relationships? I know that mixed-gender friendships can include flirting, but (again overthinking) I wonder if that would be a barrier to really close platonic friendship.
Looking at it again now, though, I see it more as a piece of street theatre and problem 2 has gone away completely. I once saw an odd. situation unfold while I was having coffee at the South Bank Centre, and it later dawned on me they were performing a show.
Another possible story would be that the women had dared each other to do this. The lesson for me is to be less lazy and try harder to work out what could be happening I didn’t like my first explanation and then it stuck in my head.
-
-
There were discussions in the forums for a week or so before when I began asking for volunteers. Three people showed some interest but Gary was the only volunteer – HE sent me the source photo to be used to make the video. Why would I use someone without consent?? Apart from the moral element there, it would probably upset a long-time and valued member of this community.
There’s no story for the test video. I was just trying out what was possible. With more work and a little imagination, there are all sorts of possibilities. Currently all the AI video sites I’ve seen are limited to 4-10 seconds per video, probably to save on workload but also, the longer the videos run, the more distorted they tend to become. I believe a thirty-second-long video would be incredibly unlikely to remain coherent and not introduce scenery and events which make no sense. Of course, I can splice together a string of 5-10 second generated videos to tell a story of sorts, or at least to explore a scenario.
What you see in that video was the third attempt. In the first, one of the AI-generated characters ran behind Gary and failed to come out the other side. In the next, two of the characters merged into one near the end. There are glitches in the one I published in this thread.
-
- 49
- Experienced Poster
Yes, it’s definitely technically interesting. Also interesting that it would still take significant human effort to make a longer story out of AI clips.
-
Unless you are super lucky, it takes some human effort to achieve anything with generative AI. Crafting the prompt takes a while (usually a few iterations) and experience helps enormously. Any details you fail to specify may come out as random nonsense. Even when you get the prompt “right”, it might take a few goes before it comes out as you want it to. I wasted most of my free startup credits because I didn’t know what I was doing. I’ve found image=>video MUCH easier to control than text=>video.
I spent a surprising amount of time making my prompts mean the same while being shorter, as there’s a character count cap, so fitting everything you need to say can be tricky. You also need an awareness of how much action or movement can be achieved in the number of seconds you choose to generate.
There’s definitely a skillset to it.
Also, so I’m acutely aware that we are still essentially in the early days of this tech being available to the public. It will get better, and it’s not going away. I think of it like computers replacing typewriters, silent movies progressing to sound, broadcasting’s far-reaching effects on live productions. You don’t have to like the technology, and where it is now may not be the same as where it goes over the next few years, but it’s here and we either get involved or get left behind.
-
I can 100% confirm my consent!
-
-
- 23
- Poster
The bigger issues with AI is the computational power required to be more precise. I have stablediffusion installed for photo creation. You change so many settings to make it better. You can change how many steps it does over the image to make it batter and how much it listens to the prompt and tries to get exact with the cfg setting. The issue is the more iterations you do and the higher the threshold on the on the cfg to pay more attention to the prompt it makes it exponentially longer to generate anything.
It’s not that it’s not clever enough now it absolutely is it’s just with the perfect prompt on a free or cheap service then you arent likely to get a great image/video.
The other thing with all these AI sites is that you basically feed the model every time you use them. Its why companies are very clear you should never upload anything sensitive to them.
-
- 23
- Poster
I spent a while on pollo ai looking for answers but couldn’t find any. The thing I found using search engines though. Thebaincomes up with the following
When images are uploaded to AI editing websites or services, they may be used to train or fine-tune the AI model, even if the user doesn’t explicitly consent. This means that user-submitted data, including personal photos, can become part of the AI’s training dataset. However, if you’re training a model on your own computer, your images won’t be shared unless you choose to upload them.
Here’s a more detailed breakdown:
AI Training:
AI systems, including image generators, rely on vast amounts of data to learn and improve. This data often includes images uploaded by users.
Data Usage:
When you upload images to an AI editing website, it’s likely that the service will use that data to train its model, even if it’s not explicitly stated.
User Responsibility:
It’s crucial to be aware of the potential for your images to be used for training when using AI tools.
Just saying be careful about uploading content you might not to end up being used in other people’s works. And check the policy of the sites you use.
-
I can’t find a downloadable one that doesn’t require a discrete graphics card (and/or Linux). It would be good to not need credits to use it.
I’ve always assumed that any photo put online will end up in all the AIs training data. I know a huge number of the photos I’ve put online are copied and posted in MANY places. It’s illegal, but what can I do?
-
- 23
- Poster
Yeah a decent GPU is a requirement. Trying to run these on CPU only resources would probably allow you to cook a full fry up on your pc. That CPU is gonna get hot. I might get comfyui installed tomorrow and see how it goes. I have a pretty decent GPU. But even then people say a short video takes a long long time.
-
I’m on a laptop so adding a suitable GPU isn’t an option.
The code used for LLMs is madly inefficient right now. We should see 8x improvement once someone optimises it, that combined with processor improvements in the next few years should see impressive speed advances. I’m assuming the next generation of main processors will have GPU-like qualities ready for this sort of processing too.
-
- 23
- Poster
I downloaded comfyui and a few different models and very quickly found my 30 series GPU struggles with many of the models. Got a couple to work a bit but need to make some PC changes before i can run the better models. Will keep trying to find something that works.
-
@Ed
“In terms of getting what you specify. Some are better than others. A few of the ones which do not permit nudity are very accurate on numbers, genders, ages etc.”
I agree, the one that I use most is meta AI because unlike a lot of AI’s it seems to realize that people don’t have six fingers and three legs! A lot of the ones that do allow erotic imagery tend to be more likely to make those errors, like something will seem fine but then on closer examination when I was illustrating one of my stories somebody pointed out why does the woman have three arms and I didn’t even realize that until somebody pointed it out!
so yeah if it’s not for anything explicit I tend to just use meta, when it comes to stuff that might include nudity I use the getimg.AI website that I think you told me about in the first place but that one now they have reduced the number of pictures you have per month where it used to be like you could get like 100 now you get 140 but each picture cost you five credits so that really brings it down to like less than 30 pictures a month. One that I use more often now is Seaart which gives you 130 credits a day and each picture will cost you may be like six credits. I think increasingly for erotic content that that one is the best one but I find it almost none of the ones that I have used will do male nudity, they will almost always put pants on the man, which is why when I illustrate my nudity stories stories about female nudity get lots of illustrations but male nudity I don’t usually get many good illustrations.
I do find though that the more complex instructions you give them the more mistakes they tend to make and you are right that almost all AI seems not to be very good with counting where I will tell it to give me an image with two men and two women and it will give me several images one with three men and one woman, one with all men, one with all women or something like that so it really gives you exactly what you want. Basically every tim
-
I haven’t tried it recently, but I’ve been told Perchance took a major step backwards in quality last year. I suspected they turned down the processing available for each prompt.
I haven’t found any AIs are any good for writing fiction. You see phrase patterns, odd grammar and poor descriptive word choices, repeated over and over. Any experienced writer will spot them very quickly, and that’s before the continuity errors and poor understanding of physics and hierarchies (He removed his hat, revealing his T-shirt etc.).
All of the ones I’ve tried are better now than last year, BUT there is a storm coming. A lot of the stuff they are learning from is AI generated, so these mistakes and faults are being fed back into the system. This is a feedback loop of junk in the system and I can’t see how it won’t just cause it all to deteriorate again.
-
-
I did find one other issue with the perchance story writer is that when I am writing naked stories sometimes it seems to forget the detail of the fact that the character is naked because I will have a character who is naked and then all of a sudden it will say that they were adjusting their clothing which they couldn’t do if they were naked! So it does sometimes make mistakes like that so you have to really watch carefully when you are doing these to make sure you catch and correct any errors that it might make. When you are using AI for anything I think you have to carefully edit and examine it after the fact because AI does still make a lot of errors that you might not always catch right off the bat, sometimes they can be fairly subtle like that.
-
@Ed
“I haven’t tried it recently, but I’ve been told Perchance took a major step backwards in quality last year. I suspected they turned down the processing available for each prompt.
I haven’t found any AIs are any good for writing fiction. You see phrase patterns, odd grammar and poor descriptive word choices, repeated over and over. Any experienced writer will spot them very quickly, and that’s before the continuity errors and poor understanding of physics and hierarchies (He removed his hat, revealing his T-shirt etc.).
All of the ones I’ve tried are better now than last year, BUT there is a storm coming. A lot of the stuff they are learning from is AI generated, so these mistakes and faults are being fed back into the system. This is a feedback loop of junk in the system and I can’t see how it won’t just cause it all to deteriorate again.”
I only started using this in the last month or two and I find it helps me to write more quickly but you are right there are certain errors. Also it uses the same names for characters over and over again if you don’t specify character names when you start the story. I am like dear Lord there are more names for women than Jenny and Rachel, and more mail names than Mark, Larry and Alex. So it does have a lot of repetition so those are things that you often have to correct. You are right they do tend to use certain phrases over and over again, and they do tend to make errors like despite the fact that is established that my character is naked they end up adjusting their clothing. So you have to be careful and make sure to catch these errors and correct them right away.
I like I said it sometimes seems to try and move things towards a reasonable conclusion like everyone was happy and they resolve this etc. etc. where my stories tend on the Darkside of things so it’s like yeah some of the stuff that it keeps going towards gets a little bit repetitive. And it’s also true that when you are trying to end the story and tell you specif
-
- 49
- Experienced Poster
On Youtube there’s a video called “The Peter & Bec Podcast ep 155”. From around 22mins 36seconds, Peter Coulson is giving his views on AI in the photography/video industry. He’s an eminent photographer, so you can imagine his take on the subject!
-
I wonder if anyone has noticed but I tried illustrating one of my naked stories today using the get image AI website and although it looks like it now gives you daily credits instead of monthly ones, but every time I try to make a nude picture it said restricted content try a different prompt. On the other hand I finally found that hugging faces working again but it doesn’t always give you naked even when you ask for naked. How about everyone else?
-
Like most of the generators, you have to use stress and repetition to get reliable full nudity. On HuggingFace, it depends very much on which model you use.
I’m going to spend more time experimenting. It would be good to find a good one with a 20-30 second time limit instead of the 4-5 seconds most enforce.
-
-
Yes, that’s about my assessment of it too. It does much more than most of us would have predicted could be achieved even a couple of years ago, but it still has a very long way to go.
As it presently stands, AI would be better described as Superficial Intelligence rather than Artificial Intelligence. If we asked a school kid to draw a picture of 4 men and 3 women seated alternately M-F-M-F at a round table, and the kid drew 6 men on the left and 3 women on the right, we would rightly declare that kid to be an idiot. But that’s precisely what AI does at least half of the time. What it does, it does well, but it rarely does exactly what is requested.
But it’s not just images that AI can’t cope with. Amazon has a chat bot called Rufus, which is presumably running some kind of AI behind the scenes.
- I asked it what battery the product I was looking at uses.
- Idiot Rufus responded without fully reading my question: “the product is powered by a battery” or something very similar.
- I asked again “but what kind of battery is it, and is it rechargeable”.
- Idiot Rufus responded “Sorry, I have checked again. The product is hand powered”.
- I asked again “One of the previous buyers said they had to replace the batteries before they could get it to work”
- Idiot Rufus responded “There are no batteries. The product is hand powered”.
- I looked at the image. It has a sticker on it which clearly says “battery powered”. I told Idiot Rufus that, but it still insisted that the product was hand powered.
- There seemed to be no way to convince Idiot Rufus that it was wrong, so I gave up.
- The product has arrived, and needless to say, it does have batteries — 4 x AA, non rechargeable.
The problem is that intelligent businesses such as Amazon are being hoodwinked into using AI for purposes that it is not yet capable of doing reliably, and making the business look as stupid as its incompetent AI chatbots.
-
That’s where companies are jumping on a bandwagon because they don’t want their competitors to leave them behind, but the people at the top do not understand the current limitations of the technology. Also, a lot of tech is using the term AI when it’s nothing like that.
I hate the use of the term AI but what I think of as AI is now called AGI, so I’m just out of date.
In terms of getting what you specify. Some are better than others. A few of the ones which do not permit nudity are very accurate on numbers, genders, ages etc.
The biggest thing is the rate of progress. Give it another 2-3 years and what we’re seeing now will be completely left behind. One thing they really can’t do is write prose. Writing at the level of producing notes or creating an e-mail is pretty good, but when it comes to literature, not even close yet.
-
In that case, the AI can only be as good as the description. Inaccurate, incomplete or contradicting descriptions and reviews are always going to thwart the desire for accuracy. Just look at some of the reviews on Amazon (and other sites) that make no sense, are wrong or simply fake. An AI has no chance.
-
- You must be logged in to reply to this topic.