AI FLM thread

[Rich_Visiting]

I've been toying with AI for a while now. But still I'm a complete NOOB and getting nowhere lol. But the potential is crazy.

I want to move away from these "apps" and actually use a standalone PC program (offline).

Any recommendations welcome.


This is a thread discussing the best AI software for editing video / images ONLY. Do NOT post images/videos or save them on your computer, never store/save anything.

[mgray]

Here is some general advice: I use automatic1111 for image generation and there are google Colab for dreambooth/model training both have near infinite walkthroughs on youtube.

The Ai tech is changing and growing almost daily but it's at a decent place right now for image generation of solo people.

You can use sites like civtai https://civitai.com/ for pretrained models for example a popular one is realisticvision. You can dl the lora file and place it in your Lora folder in your model directory, then go to the lora tab, click realisticvision and it will add it to your prompt. Beside the prompt name it will have a number value. If you up that value it will make it more similar to the initial images trained for the model, but if you lower it. It may just give your results a more detailed, realistic look.

Within automatic1111 you can highlight words, one example is "full body portrait". If you highlight that series if words press ctrl + up/down key it will alter the "weight" so if it is being stubborn and not giving good results you can try upping the weight to 1.3 and it may nudge it to give more accurate results. Conversely you can lower it if the prompt is working too strongly. Like freckles might appear as very bright unrealistic dots on the face and by lowering the weight they may appear more natural. I suggest looking up frequently used prompts and what they do, including negative prompts. Learning about the Asian-less negative prompt was a big one for me, when I noticed many of my results looked asian.

I generally get better results using very detailed prompts with weights and loras. You can slowly fine-tune your results altering these 3 parameters as needed. If you super fine tune your prompt you may get more accurately what your aiming for but may somewhat limit the variety of results you're generating. Also experiment with image size. I believe the normal default is 520x520 but by making the image taller you may get a more full body style picture. Ive also got heads stacked on heads and all other kinds of abominations.

Obviously if you choose to do any of this on your own pc or on a public website use caution and basic common sense. Especially on any public website I'd just assume everything i do can and will be seen.


I'm a pretty big fan of warhammer fantasy and age of sigmar(40k too but I'm a fantasy nerd through and through) here's some results of chaos warriors i was able to generate:


[Image]
[Image]
[Image]

[Sully23]

I really see very dangerous AI, we will see a lot of fakes images of some celebrities, something similar almost 15 years ago I was googling supposedly leaked pictures of Taylor Swift or Katy Perry but they turned out to be fakes.

The good thing about AI is that it generates images for artistic purposes, the bad thing is that it has its limits and you can't show too much.

[Rich_Visiting]

Here is some general advice: I use automatic1111 for image generation and there are google Colab for dreambooth/model training both have near infinite walkthroughs on youtube.

The Ai tech is changing and growing almost daily but it's at a decent place right now for image generation of solo people.

You can use sites like civtai https://civitai.com/ for pretrained models for example a popular one is realisticvision. You can dl the lora file and place it in your Lora folder in your model directory, then go to the lora tab, click realisticvision and it will add it to your prompt. Beside the prompt name it will have a number value. If you up that value it will make it more similar to the initial images trained for the model, but if you lower it. It may just give your results a more detailed, realistic look.

Within automatic1111 you can highlight words, one example is "full body portrait". If you highlight that series if words press ctrl + up/down key it will alter the "weight" so if it is being stubborn and not giving good results you can try upping the weight to 1.3 and it may nudge it to give more accurate results. Conversely you can lower it if the prompt is working too strongly. Like freckles might appear as very bright unrealistic dots on the face and by lowering the weight they may appear more natural. I suggest looking up frequently used prompts and what they do, including negative prompts. Learning about the Asian-less negative prompt was a big one for me, when I noticed many of my results looked asian.

I generally get better results using very detailed prompts with weights and loras. You can slowly fine-tune your results altering these 3 parameters as needed. If you super fine tune your prompt you may get more accurately what your aiming for but may somewhat limit the variety of results you're generating. Also experiment with image size. I believe the normal default is 520x520 but by making the image taller you may get a more full body style picture. Ive also got heads stacked on heads and all other kinds of abominations.

Obviously if you choose to do any of this on your own pc or on a public website use caution and basic common sense. Especially on any public website I'd just assume everything i do can and will be seen.


I'm a pretty big fan of warhammer fantasy and age of sigmar(40k too but I'm a fantasy nerd through and through) here's some results of chaos warriors i was able to generate:


[Image]
[Image]
[Image]

That's great, many thanks, this is really helpful and I will take a look. I've not really had a chance to see what AI can do. But I personally see it as a way to make our own worlds without harming anyone. If the results are used to harm people then that's completely different, I would never do this, and I have no sympathy with those that use it to hurt others for their own gratification.

I really see very dangerous AI, we will see a lot of fakes images of some celebrities, something similar almost 15 years ago I was googling supposedly leaked pictures of Taylor Swift or Katy Perry but they turned out to be fakes.

The good thing about AI is that it generates images for artistic purposes, the bad thing is that it has its limits and you can't show too much.

I remember a fake nude of Christina Ricci (Adam's Family) floating around in the late 90's before even AI No harm was done that day. I also remember a Hermione Granger panty shot where she is sat at a cauldron But now it has potential to produce more explicit material, but in the end its still just a set of pixels on a screen for the use in the privacy of ones own home.

[Sully23]

I remember a fake nude of Christina Ricci (Adam's Family) floating around in the late 90's before even AI No harm was done that day. I also remember a Hermione Granger panty shot where she is sat at a cauldron

Christina Ricci was my eternal crush and Wednesday, what a shame she never had her OT moment, but when I started looking for nude celebreties, I found her only topless in Prozac Nation, many don't even know this movie.

So many years waiting for a nude Emma Watson, the only thing she did was a body double and a bathtub double that would be an Off Topic here, even though she once said she wanted to get naked in some project after finishing Harry Potter.

Spoiler: Harry who did it in the theatre.

[Rich_Visiting]

I remember a fake nude of Christina Ricci (Adam's Family) floating around in the late 90's before even AI No harm was done that day. I also remember a Hermione Granger panty shot where she is sat at a cauldron

Christina Ricci was my eternal crush and Wednesday, what a shame she never had her OT moment, but when I started looking for nude celebreties, I found her only topless in Prozac Nation, many don't even know this movie.

Maybe it was a real photo such were the late 90's lol

[Rich_Visiting]

what a shame she never had her OT moment

AI says that maybe she did have that OT moment why not create it? Let's drag FLM into 2025 and see where this thread goes...

[Sully23]

what a shame she never had her OT moment

AI says that maybe she did have that OT moment why not create it? Let's drag FLM into 2025 and see where this thread goes...

Rich, the results are very unrealistic, put Miley Cyrus (recent photo) and go through the conversion and you see that it is not in accordance if we compare with some nudes of the past, they will only add more size to her topless as if we did a digital mammoplasty.

[Rich_Visiting]

what a shame she never had her OT moment

AI says that maybe she did have that OT moment why not create it? Let's drag FLM into 2025 and see where this thread goes...

Rich, the results are very unrealistic, put Miley Cyrus (recent photo) and go through the conversion and you see that it is not in accordance if we compare with some nudes of the past, they will only add more size to her topless as if we did a digital mammoplasty.

not in our own worlds yet then, it will come

[Rich_Visiting]

Rich, the results are very unrealistic,...

I'm just learning but made Mathilda (Leon) into Kick Ass scene with software on tor rent...

https://www.youtube.com/watch?v=neXG2ybgJCc

OMG, can't embed

[Image]

wow