How to Improve AI images

[RobynHood]

AI images can be amazing but they can also be really strange. I have enjoyed playing around with A.I. to where I now have a few tips and tricks I can use to get great photos.

In my journey thread 'The Great SEO Heist of 2024'
https://www.blackhatworld.com/seo/the-great-seo-heist-of-2024.1558453/post-17256288

I showed how you can use stock image sites to create great AI images like this one:



but what if you have a specific image in mind. For this second method we are going to use a variation of a prompt people are probably familiar with in A.I. writing.

We are going to use the phrase “in the style of …..”

Here’s an example prompt I used with stable diffusion:

mystical photograph of a young first century Jewish fisheman emerging from the misty waters of lake Galilee, wet hair and skin, neck deep, weathered face his expression serene and calm. The atmosphere is surreal and ethereal, with soft blue tones reminiscent of Michael Kenna's landscape photography.

and here is the result.


You will notice in the prompt I include some details of the type of photo and the subject of the video but then add a on the style of.

If you don’t know any photographers google top 100 photographers of the 21st century or something similar. See which styles you like and use it in your prompt.

I will post more prompts and examples as we go on. What tips do you have to improve images?

 

[Gants-P]

Esgran is what you want: just Google esgran colab add your image and walah.


Anyways generating images isn't the problem it's consistency over the same image creating 20 images.

Example:
Sarah was born, (10)
Sarah was a child, (15)
Sarah went to school, (50)
Sarah finished high school(100)
Sarah's life (500)

Trying to get AI to do that consistently is like asking someone to burn their hand instead of yours and still get paid.

Anyone can generate images, and prompt finding a trove of photos not so, well Facebook and Instagram sure.


Another example is giga chads Instagram. Consistent, yet no one at first had a clue if it was real until now.

 

[RobynHood]

Esgran is what you want: just Google esgran colab add your image and walah.


Anyways generating images isn't the problem it's consistency over the same image creating 20 images.

Example:
Sarah was born, (10)
Sarah was a child, (15)
Sarah went to school, (50)
Sarah finished high school(100)
Sarah's life (500)

Trying to get AI to do that consistently is like asking someone to burn their hand instead of yours and still get paid.

Anyone can generate images, and prompt finding a trove of photos not so, well Facebook and Instagram sure.


Another example is giga chads Instagram. Consistent, yet no one at first had a clue if it was real until now.

do you mean this kind of thing

 

[Gants-P]

do you mean this kind of thingView attachment 317170View attachment 317171View attachment 317172View attachment 317173

No, you'll know when you see it, first you need to train the data, I don't believe you have, if so it's on a different source.

I cannot see consistency there for the immersion is broken, that's the hard part.

 

[medialounge]

Esgran is what you want: just Google esgran colab add your image and walah.


Anyways generating images isn't the problem it's consistency over the same image creating 20 images.

Example:
Sarah was born, (10)
Sarah was a child, (15)
Sarah went to school, (50)
Sarah finished high school(100)
Sarah's life (500)

Trying to get AI to do that consistently is like asking someone to burn their hand instead of yours and still get paid.

Anyone can generate images, and prompt finding a trove of photos not so, well Facebook and Instagram sure.


Another example is giga chads Instagram. Consistent, yet no one at first had a clue if it was real until now.

It's difficult to see consistency in an AI picture that looks real.

 

[alonelywriter]

Are you using Stable diffusion?

 

[RobynHood]

No, you'll know when you see it, first you need to train the data, I don't believe you have, if so it's on a different source.

I cannot see consistency there for the immersion is broken, that's the hard part.

Do you want to add something constructive and show us what you mean and include prompts and workflow

@alonelywriter yes it’s stable diffusion. Using prompt and the same seed image.

 

[Gants-P]

Do you want to add something constructive and show us what you mean and include prompts and workflow

@alonelywriter yes it’s stable diffusion. Using prompt and the same seed image.

I'm not trying to be a ass, I'm just stating the obvious, if you are here to generate images then what I says irrelevant, if it's to create a only fans page with back links then yes consistent content is important.
(I recommend tensor art, if you have no access to a rig, however you can create consistent images but you'll need multiple email accounts to game the site) I highly recommend having a personal rig because you can batch feed images and delete the mistakes) batch feed is sending images back in for training.
Most of my images are 5mb+ I'll upload a few images to illustrate my point, so I'll list this thread as watch for the mean time.

 

[Nikolaos]

AI images can be amazing but they can also be really strange. I have enjoyed playing around with A.I. to where I now have a few tips and tricks I can use to get great photos.

In my journey thread 'The Great SEO Heist of 2024'
https://www.blackhatworld.com/seo/the-great-seo-heist-of-2024.1558453/post-17256288

I showed how you can use stock image sites to create great AI images like this one:

View attachment 317140

but what if you have a specific image in mind. For this second method we are going to use a variation of a prompt people are probably familiar with in A.I. writing.

We are going to use the phrase “in the style of …..”

Here’s an example prompt I used with stable diffusion:

mystical photograph of a young first century Jewish fisheman emerging from the misty waters of lake Galilee, wet hair and skin, neck deep, weathered face his expression serene and calm. The atmosphere is surreal and ethereal, with soft blue tones reminiscent of Michael Kenna's landscape photography.

and here is the result.
View attachment 317141

You will notice in the prompt I include some details of the type of photo and the subject of the video but then add a on the style of.

If you don’t know any photographers google top 100 photographers of the 21st century or something similar. See which styles you like and use it in your prompt.

I will post more prompts and examples as we go on. What tips do you have to improve images?

Truly magnificent pictures. I wouldn't tell if the image is real or fake.

 

[pbnpanda]

The remarkable pictures. It's challenging for me to discern whether the images are authentic or manipulated

 

[frozen_yoda]

Truly magnificent pictures. I wouldn't tell if the image is real or fake.

I think it's a good effort. Maybe from a phone it would be harder to tell. On my computer it's pretty obvious though.

If you're just scrolling and not taking a close look you might think they're real. Number 3 you might just think she has a disability.

The most passable is number 2.

-----

It's definitely going to continue to improve over time. For now, whether it's yourself or a PA, you can produce a large amount and have them sift through to find the best ones.

"If you get in once it's perfect, you're in too late."
- FY

 

[RobynHood]

Tip 2 camera and lenses

Other ways to add realism to you images is to use lenses in your prompt like the Sigma 24 mm for the focal length or OM system 12–40mm PRO II 40mm, 1/100 sec, f/2 8, ISO 800

Or use a specific camera like Fujifilm XT3

Research up on the effect you want and you can get some great images using the right camera and lens.

To add an extra dimension including light information helps as well