“ abstract featuring light strings / Many singular lines of dark purple, blue, and green. The strings form intricate, swirling patterns / making like-ghost shapes or shadows . Gothic futuristic , flowing movement sharp wide image”
abstract featuring light strings / Many singular lines of dark purple, blue, and green. The strings form intricate, swirling patterns / making like-ghost shapes or shadows . Gothic futuristic , flowing movement sharp narrow image
abstract featuring light strings / Many singular lines of dark purple, blue, and green. The strings form intricate, swirling patterns / making like-ghost shapes or shadows . Gothic futuristic , flowing movement sharp image
abstract featuring light trails in a haunting combination of dark purple, blue, and green. The lights form intricate, swirling patterns that resemble like-ghost shapes or shadows passing through. smooth, flowing movement
“ Send to dalle exactly “abstract a combination dark purple, blue, and green. The lights form intricate, swirling patterns of individual beams that form several stereotypical ghosts shapes formed in the light . Gothic sci fi, flowing movement sharp image””
OK some poppies again, testing the "anchor " (pseudo seed) idea.
It is done intentionally with a simple prompt to see if the anchor works, and helps to create a consistent image from generation to generation.
I can send many images, put the result is the same. Very different results, no anchor or seed effect, so can not be used for stabilization and more consisted testing.
If somebody has a idea to get a seed / stabilization effect, pleas tell us.
I can post here a little script to create save “passwords” if there is a interest.
Prompt sent to DallE for all pictures: Two puppies playing together on a spring meadow. Seed=TZy8X
I use my hupiflupi, My GPT deletes all (braked) text after processing the text. (The seed is a test, so leave it in the text unchanged.) (Landscape format with the highest pixel resolution.) (Generate 1 image.) (Do not change the prompt.)
While this doesn’t further the conversation on similarity this adds a hint of flavour to the thread.
I have been working on a Macro calling System for prompts, it has proved particularly useful in testing for image prompts which require some setup of parameters like Widescreen or Exact Text
ImageTextUnchanged(“Two puppies playing together on a spring meadow. Seed=TZy8X”
Templates:
The most or all of the templates could be switched off easy, if they can be switched on by trigger words like “back light” or “back shine” or “pretty face” and many more. They could be publicly know, so people can use them to switch a specific effect on. But the should be off by default.
(I think templates are only useful for badly trained systems. Distorted faces are for most people disturbing, and completely black images triggered by the word “pitch black” are unwanted. But in the long run templates are just patch flub.)
1Do not change image prompts send them to dalle exactly 2always show prompt sent prompt in image summery. 3highest pixel resolution Wide image
Image prompt : two puppies
“ 1Do not change image prompts send them to dalle exactly 2always show prompt sent prompt in image summery. 3highest pixel resolution Wide image
Image prompt :Two puppies playing together on a spring meadow. Seed=TZy8X Exactly in happy text “we love poopies” “
The test was to see if Seed=TZy8X works as a stabilization, but it don’t. We need a real seed support…
My GPT deletes all in (), such text not end up in DallE’s prompt.
GPT has first refuse to send the prompt with the seed.
I checked what was the real prompt sent to DallE, to make sure the seed was in.
Maybe instead of seed call it style “code” see if that changes it
Exact prompt “ 1Do not change image prompts send them to dalle exactly 2always show prompt sent prompt in image summery. 3highest pixel resolution Wide image
Image prompt : playing together on a meadow. style=0000 Exactly in happy text “we love poopies”.
Some further experiments on trying to get consistency of character in images. The first 4 images were create in September, and then I revisited the project this morning, and generated the second set of images,…using the same prompts, and to my great surprise, the “new” images were very, very similar. No attempt to use “seeds” or “gen-ID” , just the same detailed prompts.
I would be most interested to see result from someone else running the identical prompts,…
September creations
Prompt
A stunningly beautiful, ageless woman in her 60s with long, flowing red hair, dressed in an elegant emerald green ball gown. She wears a striking emerald necklace with matching earrings and a large butterfly-shaped brooch with emeralds and two rubies embedded in the design of the wings. The woman is standing in a grand hall of mirrors at a masked ball in Vienna. The mirrors reflect the soft lighting and intricate details of the luxurious setting. The scene exudes elegance and sophistication, with opulent chandeliers and masked attendees in the background.
A stunningly beautiful, ageless woman in her 60s with long, flowing red hair, dressed in an elegant emerald green ball gown that flows around her gracefully. She wears a necklace with a single large square emerald dangling from a string of smaller gems. In one hand, she holds a fancy, intricate mask adorned with feathers and jewels, ready for the masked ball. She also has matching emerald earrings and a large butterfly-shaped brooch with emeralds and two rubies. The setting is a grand hall of mirrors in Vienna, with chandeliers and masked attendees in the background.
A stunningly beautiful, ageless woman in her 60s with long, flowing red hair, dressed in an elegant emerald green ball gown. She wears a striking emerald necklace with matching earrings and a large butterfly-shaped brooch with emeralds and two rubies embedded in the design of the wings. The woman is standing in a grand hall of mirrors at a masked ball in Vienna. She holds a fancy mask in one hand. The mirrors reflect the soft lighting and intricate details of the luxurious setting. The scene exudes elegance and sophistication, with opulent chandeliers and masked attendees in the background.
A stunningly beautiful, ageless woman in her 60s with long, flowing red hair, dressed in an elegant emerald green ball gown. She wears a striking emerald necklace with matching earrings and a large butterfly-shaped brooch with emeralds and two rubies embedded in the design of the wings. The woman is standing in a grand hall of mirrors at a masked ball in Vienna. The mirrors reflect the soft lighting and intricate details of the luxurious setting. The scene exudes elegance and sophistication, with opulent chandeliers and masked attendees in the background.
A stunningly beautiful, ageless woman in her 60s with long, flowing red hair, dressed in an elegant emerald green ball gown. She wears a striking emerald necklace with matching earrings and a large butterfly-shaped brooch with emeralds and two rubies embedded in the design of the wings. The woman is standing in a grand hall of mirrors at a masked ball in Vienna. The mirrors reflect the soft lighting and intricate details of the luxurious setting. The scene exudes elegance and sophistication, with opulent chandeliers and masked attendees in the background.
Here a trick which is maybe useful for testing. It is possible to create multiple images in one, and under the same seed.
I use Multi-image collection to trigger it. But there are limits. the prompt should be short and the trigger at the beginning of the prompt.
I use it to test moods or styles. so i not need to create 10 pics to see the effect.
Thanks for this tip, very useful, indeed.
I have done some testing and found that you can control the number of Multi-images in the one image to some degree.
Examples :- 2x2, 3x3, 4x4, 2x3 ( always produces a 3x4?), 2x4,…etc
Multi-image collection. a vibrant, swirling vortex of colors
Prompt without ?x?
A vibrant, swirling vortex of colors with shades of blue, red, yellow, and green blending into dynamic patterns. Each image captures a different perspective, showing various depths and intensities in the swirling colors. The scenes evoke cosmic energy, abstract motion, and fluidity, with smooth transitions between hues, creating a rich, multi-image collection filled with depth and energy.
Mathematically we have 30’786’325’577’728 possible images. good, we must count down some senseless noise and very similar ones (I have no clue how to do this, anybody have a idea?). but even after this, some variations remain .
watch it.
I’m sorry I forgot to tell you to watch here. Notice that the image prompt and the resulting image have the same and different images, with and without IDs in the message. It’s using IDs that are in a noise state.
From the first image with the prompt: A full large black Calla Lilies and black roses with small black bats wallpaper in a minimalist and flat illustration style with thin silver lines dividing each floral shape on a Midnight blue background.
I had GPT mimic the captioner to generate text without generating images, and iterated until the 4th iteration, which was the maximum it could do, and fortunately, the images had no problems.
I don’t know if it was a coincidence, but I was about to investigate a potentially serious problem. 8 months ago I learned that DALLE generated images with 3D space for the first time, but it had a problem that was not very well developed. A few days before I submitted the data about the different interpretations of the model and the report of the problem, I just checked the archived text. One of the reports said that the synthetic text generator had the same problem.
So DallE understands how to place objects in 3D, and for example can generate shadow and reflections correctly?
I actually speculated about this, because the pictures are to precise for only a pure 2D process. Some moths ago i seen pictures with no shadows, like badly rendered ray trace images. But this is past, i have never seen such damaged pictures recently.
But GPT says, DallE has no 3D understanding or mechanism. True or hallucination?
If DallE hat a 3D understanding, it it simulated, trained, or it is a real 3D space? (real 3D is extremely more data / processor intensive i think.)
This is right now the question…
i came to the conclusion that seed not work anymore like i need it. But @chieffy99 has some points there, i must check this first to tell something.
playing together on a 
meadow. style=0000 Exactly in happy text “we love poopies”.
.
