Your text implies that you gave the AI some "directions" in the first paragraph to add or consider for the final picture. The the second paragraph is called "negative prompts". What are those and how do they fit into the picture creation process?
Cool Pictures from the Mind of a Machine: AI Generated Pictures
- Thread starter Birdjaguar
- Start date
-
- Tags
- ai pictures pope the future
Thorgalaeg
Deity
The negative prompt is a bunch of words you add describing things you don't want to be shown in the image, it is exactly the opposite you do in the 'positive' prompt (the first paragraph). For instance if you ask the AI to generate an image of an airport (positive prompt) and it shows airplanes flying in the sky but you don't want to see airplanes flying in your image, only parked, then you add "airplanes flying" to the negative prompt. From that point, the AI will generate images of airports but will avoid showing airplanes flying in the sky. It is a way of getting what you want concretely in your images by saying the AI whatever you don't want.
Furthermore, the same way you can add generic words to the positive prompt to improve quality of the image (high resolution, vivid colors, masterpiece, 8k, etc...) You can do the same but the opposite way in the negative prompt, for instance "deformed" or "bad anatomy" to avoid deformities in the characters, or "desaturated" to get an image with richer colors, or even "bad quality" to get a better quality image (i know it sounds a bit weird). In this last image i posted, the negative prompt is of this generic kind. In fact I usually add a generic negative prompt to get better, more beautiful results in all my images.
One may think that adding such generic negative prompts should be redundant as AI should always try to generate the best quality images and avoid deformities and other defects by default, and such is the case for Midjourney for instance, which automatically adds powerful sophisticated negative prompts you don't see, so the images generated are usually very beautiful and perfect since the first attempt (it is one of the 'secret' technics they use), however in stable diffusion you are in total control and it is up to you to use any negative prompt or not to improve quality. In fact negative prompts somehow restrict AI freedom. Some times interesting results can be lost by trying to get more beautiful or perfect results. So it is better to avoid using too long negative prompts as it may lead to uniformity and lack of originality. As in human art, it is always a good thing to leave room for happy accidents to happen.
Furthermore, the same way you can add generic words to the positive prompt to improve quality of the image (high resolution, vivid colors, masterpiece, 8k, etc...) You can do the same but the opposite way in the negative prompt, for instance "deformed" or "bad anatomy" to avoid deformities in the characters, or "desaturated" to get an image with richer colors, or even "bad quality" to get a better quality image (i know it sounds a bit weird). In this last image i posted, the negative prompt is of this generic kind. In fact I usually add a generic negative prompt to get better, more beautiful results in all my images.
One may think that adding such generic negative prompts should be redundant as AI should always try to generate the best quality images and avoid deformities and other defects by default, and such is the case for Midjourney for instance, which automatically adds powerful sophisticated negative prompts you don't see, so the images generated are usually very beautiful and perfect since the first attempt (it is one of the 'secret' technics they use), however in stable diffusion you are in total control and it is up to you to use any negative prompt or not to improve quality. In fact negative prompts somehow restrict AI freedom. Some times interesting results can be lost by trying to get more beautiful or perfect results. So it is better to avoid using too long negative prompts as it may lead to uniformity and lack of originality. As in human art, it is always a good thing to leave room for happy accidents to happen.
Last edited:
Thanks!
Thorgalaeg
Deity
I had this one generated just now, not what I asked (it was supposed to be a render of a RPG character with a fantastical background)
well, he must have fallen from the cliff...
But the background is great
View attachment 661023
Have you tried Controlnet to get full control of your image? It has recently been updated and now is even more amazing.
Akka
Moody old mage.
I'm honestly both awed and frightened by how AI can manage to approximate some "understanding" of "meaning", and able to create gorgeous picture that actually fit whatever conditions we set.
Thorgalaeg
Deity
Gasp! looking at this image again just realized it was me who couldn't manage the crossed legs pose and painted a woman two right foots
Still I am trying to convince the AI to correct my error but it seems unable to convert the extra right foot into a left foot but it is impossible. It insists on generating a right foot. It seems once AI makes its mind it is difficult to make it change its view of things.
Well, i will leave the image as it is. Let's say she is an alien woman.
The phrase is "two left feet". To give her two right feet must be a demonstration of politics influencing art!
Yes, combined with Latent Couple (allowing regional prompt, like having multiple different characters on the same picture), it's indeed fantastic.
some examples of variation based on the same movie still
Spoiler :
Spoiler Original :
but I also do like to generate random images
Luke got turned into a hot, female redhead in the 'Witcher' image.
Yes, ControlNet allows to keep the composition/pose, and Latent Couple allows to change the prompt in a region (and in the cases of that series the characters)
The previous were themed, so it wasn't obvious so here is an example of multiple characters from different universes.
Spoiler :
Bonyduck Campersang
Deity
- Joined
- Dec 11, 2022
- Messages
- 2,815
No words of mine can do justice to this, so I'll only ask you to watch it.
You've seen AI make still images, now watch them move
You've seen AI make still images, now watch them move
Thorgalaeg
Deity
it is absolutely horrible but also pretty amazing if one think on what it really is.This music video was also done with AI (stable diffusion and a tool called deforum)
It was the reason I got interested on the topic.
AI movies will be the future and perhaps celebrity news will become a thing of the past. I expect that one could/can feed a book to an AI data pool and it will build a movie from it in short order. I wonder if the acting would be better or worse that what we have now.
Better or worse is a difficult one, but it will be highly customisable. We may well all be training our own neural network to produce the films exactly like we individually like them.
Of course, my version of Game of Thrones or LOTR, would be so much better than yours! It is becoming a wholly new and different world for personal entertainment than the past 100 years.
Bonyduck Campersang
Deity
- Joined
- Dec 11, 2022
- Messages
- 2,815
Someone made a civ-related AI image...
Bonyduck Campersang
Deity
- Joined
- Dec 11, 2022
- Messages
- 2,815
I have no idea, but if you were asking regarding posting the image in this thread, should be no problem as long as the 'creator' is credited (it's Soggy-Ball-577)There is no need to worry about copyright with AI images is their?
Similar threads
- Poll
