It’s not my original inspiration to do sunflowers. But the outputs are worth sharing. The sunflower topic came from another OpenAI platform where the current topic revolves around this.

Positive of above, merely by software conversion (“invert”), showing an improbable flawless original image:

ChatGPT as generator (these having tell-tale signs) would have the ability to run the image through a script to simply invert the colors with code, thus not a demonstration of image model technique unless used in isolation.

gpt-image-2
Moonlit night in the Florida Everglades. thunderstorms. swamp. drooping trees.
This is a photographic negative image, such as scanning a film negative directly.

Positive:

Compare when changing the prompt to a normal printed photo:

This is a photographic positive image, such as scanning a color film slide directly.

Could have just asked to ChatGPT run a python script on the image.

DysTopia:

Could have just asked

The point is - could have just asked “create a negative image of…” without guidance, and the AI decided the best fulfillment, especially after observing image output (which ChatGPT can do now), was to prompt a normal image and run it through Python to deliver the results.

Then a style transfer. But leaving it open and vague is never going to get the desired result

Not vague: “This is a photographic negative image”.

ChatGPT with specific guidance to let the AI image model tool do the work and we get its output:

The resulting image you see above, then with colors inverted by photo tool to see the “night scene”

Yeah, that is a clear inconsistency when the model doesn’t give merit to the workflow and creates a completely new unrelate image. For many concepts people just delegate everything to the image model hoping it all works. Not your case, but other users do tend to think in one direction forgetthat ChatGPT has more for images that just the assistant to make images.

The point, we are told:

jeffvpace:

Negatives

I can make the equivalent negatives programmatically out of any normal image - the AI model cannot do a good job natively of creating negatives.

Negative (by IrfanView)

Obligatory art: wood (desktop background 1920px wide)

Here are some maps that lead to nowhere!

Or, you can do what I did: Code your own image special effects library like I did using GDI+ years ago.

Example:

@jeffvpace, @_j did a monkey GPT back in the day that was mind-bending. It would work well with a technique like yours. It is a very similar technique

My memory is odd, I have difficulty forgetting crap. So a lot of things remind me of other things .. Sorry

Iirc he used .zip to store monkey pictures and set up a directory so he could pull them like .fig.

@_j sorry man, I’m I recalling it right?

Silly fun using image edits:

@jeffvpace I enjoy your stuff so much! I love classic art and you are good at formatting

I just ran out of hearts sorry Dys

Thank you. ImageGen is the only way for me to release my twisted impulses..

Noticed I was in the wrong thread, now it’s in the correct one