• Welcome to the forums, Guest. Please note that you must make a post in the introduction thread and upload an avatar to gain full access to the forums.

Created AI pics for OLD, looking for some advice

First of all I am amazed how you create sharp backgrounds with AI for portraits? I struggle exactly with this when using Flux as a base model.

Other then that I wouldn't recommend it. I tested it myself and the results are similar to the real photos. But the main problem is if you are going to a date and then need you to explain the photo situations.

And some images are easily detectable that they are not real:

IMG_20241108_100516.png

IMG_20241108_100451.jpg


So you need to be very careful for looking any potential issue in the photo.
 
Please don't do this.
Why do you say this? Is it because the photos are obviously fake? or just the principle of it?

For me, it probably would have cost thousands of dollars in photoshoots and an insane amount of time spent just to get 2 or 3 photos of this quality. With AI, I was able to do it in 4 hours. I think this is the just modern age and what we need to do for dating apps.
 
Some pics are obvious AI, others look realistic. Holy shit, AI is getting better every day. How did you do that?

Would it be possible to refresh real pics of yourself with AI? Better outfit or something like that?
Somehow the stuff scares me, anyone can have high level pics now.

Most people won't get it anyway. I know an IG account of a guy who apparently uses the maximum FaceAPP filter. I even think he takes the maximum filter, saves the image and slaps the filter over it again.
It's so extremely obvious that the picture is fake and yet women are simping for this account.
I just followed a Youtube video I found, not sure I can link it here but if you want I could DM you it.

I don't think so, you basically give the model an image prompt and it creates one based off of that. Not sure how I could upload a real image of myself and try to upgrade that specific image.

I agree, just need to be careful don't wanna be shadow banned for posting fake pics haha.
 
First of all I am amazed how you create sharp backgrounds with AI for portraits? I struggle exactly with this when using Flux as a base model.

Other then that I wouldn't recommend it. I tested it myself and the results are similar to the real photos. But the main problem is if you are going to a date and then need you to explain the photo situations.

And some images are easily detectable that they are not real:

View attachment 1016

View attachment 1017


So you need to be very careful for looking any potential issue in the photo.
There's a very specific prompt format I found that helps create the lighting and background for the flux model. I can share it with you if you want.

Yeah for whatever reason the model was really fucking up my legs and feet lol. I think I didn't share enough full body pics in the training lora model. I'll probably just crop the pics closely to make it less liable to AI detection.

Hmm that is true, I'm not a huge fan of lying on dates with girls. I might use a couple AI photos and a couple of my personal real photos and see how it goes.
 
There's a very specific prompt format I found that helps create the lighting and background for the flux model. I can share it with you if you want.

Yeah for whatever reason the model was really fucking up my legs and feet lol. I think I didn't share enough full body pics in the training lora model. I'll probably just crop the pics closely to make it less liable to AI detection.

Hmm that is true, I'm not a huge fan of lying on dates with girls. I might use a couple AI photos and a couple of my personal real photos and see how it goes.
If you can share the prompt or DM it to me that would be cool.
 
I'm impressed it got the hands and feet mostly right. Still got some issues with textures, skin and clothes, gotta watch out for that. Also, I think you're not seeing big enough: why wouldn't you make pictures you'd have a hard time getting in real life, like you speaking at a business event, you smoking a cigar with friends at a fancy bar, or something like the one with the dog (tough one though, AI still got bit of difficulty with animals)?

this is the future boyos
As soon as enough guys start doing this, you bet match.com and company will start pouring money into complex AI-detection tools.
 
I'm impressed it got the hands and feet mostly right. Still got some issues with textures, skin and clothes, gotta watch out for that. Also, I think you're not seeing big enough: why wouldn't you make pictures you'd have a hard time getting in real life, like you speaking at a business event, you smoking a cigar with friends at a fancy bar, or something like the one with the dog (tough one though, AI still got bit of difficulty with animals)?


As soon as enough guys start doing this, you bet match.com and company will start pouring money into complex AI-detection tools.
Yeah they will, but y'all need to find a way to win before it comes to that.

What I saw this year was, profiles that were top 10%, no longer got jack sh*t. Or got horrible quality.

It's top 1-5% or death buddy boyos, OR have the correct archetype/phenotype which is geentics. So, what choice do we have? I'd lean into ANYTHING that is working and rinse the shit out of it.

I would have accepted the don't open the AI box of worms last year, I didn't touch this shit myself. After this year, and seeing how bad it was out there, there is no justification not do do whatever it takes to win if you're doing online.
 
I'm impressed it got the hands and feet mostly right. Still got some issues with textures, skin and clothes, gotta watch out for that. Also, I think you're not seeing big enough: why wouldn't you make pictures you'd have a hard time getting in real life, like you speaking at a business event, you smoking a cigar with friends at a fancy bar, or something like the one with the dog (tough one though, AI still got bit of difficulty with animals)?


As soon as enough guys start doing this, you bet match.com and company will start pouring money into complex AI-detection tools.
The AI still isn't very good with props or having people in the background. The model is only trained on my likeness so I can't produce anyone else in my pics. There's still a ways to go there.
 
I tested your approach with the prompt generator. In my case it still generates a shallow depth of field. I think this happens because I trained the Lora on images with a shallow depth of field. Just for fun I need to test it with lower quality images made from a phone
 
I tested your approach with the prompt generator. In my case it still generates a shallow depth of field. I think this happens because I trained the Lora on images with a shallow depth of field. Just for fun I need to test it with lower quality images made from a phone
Yeah I trained my lora on basically all phone selfies and a couple full body shots.
 
Back
Top