I’m out of hearts but I’ll be back. All your work is lovely folks. 
6 Likes
Your Bells remind me of a certain fuzziness of a kitten wondering what the difference is between emotio and ratio 
Another experiment
4 Likes
Attempt to reconcile elements close to Catholicism and transhumanism in an image. The idea behind the composition of the image is not very original, and I suppose it could be improved. Do you think the theme I mentioned is worth exploring artistically, or does it risk coming across as banal and needlessly provocative?
4 Likes
I think it’s perfect for artistic contemplation 
In this context:
The Bible says that humans are created in the image of God.
Well, so it is natural that humans are also a creating being
2 Likes
I’m currently having fun testing something.
I’m giving it images of smiles, foreheads, noses… everything I find aesthetically pleasing. I ask it to analyze and look for patterns to see what I naturally like. Then I ask it to create an image of the perfect face based on the descriptions. And, overall, it doesn’t disappoint.
I’ll need to test it more. The goal was to create a police-style composite sketch, but it’s far from that.
Still, it might help me create a personalized avatar in the future, haha.
At first, I thought I could see what kind of improvements I could get with plastic surgery tweaks, lol, but no, it’s not useful for that either.
If anyone has suggestions for creating the most realistic faces possible, let me know, please!
3 Likes
They look quite good in terms of facial structure, the suits, etc.
To give your some poissble tips:
Reiterate the pics somewhat more.
Because: Mostly DALL-E 3 is making better pics with some sunlights or tell it to add additional light sources, like in a photo studio, etc. then you may be able to get 100% real looking faces, with skin pores etc. I hope this makes sense.
Sometimes those pics look too perfect, this makes humans then appear to look very plastic like. Like a barbie doll or so.
Then also we perceive them as way too unreal.
So, tell it to weave some inconsistencies, impurities in. If that makes sense.
But hey, it’s starting to look good. 
You already went quite far. You’ll get there.
2 Likes
ok, now think, additionally: What may a photographer do?
They may possibly adjust the lighting somewhat, give it a higher resolution (make parts of the iris maybe possible, when zooming in closer?) The pores are already visible, mabye the head turned slightly or making them look over their shoulder. Etc.
Or when front shots then you may want it to turn out like this, I guess:
EDIT: I guess you’re going for someone like this, probably, aren’t you?
3 Likes
Koreans always have amazing skin, so AI-generated images are less noticeable, hahaha. They come with a built-in filter in real life 
1 Like
That’s for sure. I’ve met quite a lot because of a good friend who’s also into Koreans.
So, here’s another one:
Then you could vary this now yourself.
Prompt below (non layered):
hyper-realistic portrait of a Korean woman, focusing on refined skin textures with fine pores, faint freckles, and natural pigmentation for a lifelike appearance. The eyes feature detailed reflections of soft light sources, with realistic tear ducts and a warm, expressive gaze enhanced by subtle iris details. The hair is finely detailed, showing individual strands, slight frizz, and dynamic highlights that interact with the lighting, cascading naturally in soft waves. The lighting is cinematic, emphasizing facial contours with balanced warmth and shadow, while introducing ambient light for depth on the neck and jawline. The background features a softly textured gradient in a neutral gray tone with enhanced bokeh effects to keep the subject in focus. The lips are refined with natural creases and a soft gloss that reflects light for a realistic hydrated look. Faint fabric textures and light shadows are introduced on the visible shoulders to avoid flatness and complete the polished aesthetic. The overall image is photorealistic, blending warmth, texture, and subtle emotion for maximum realism.
EDIT: This is after two iterations. since I’ve enhanced the workflow now.
3 Likes
@crunchybomb07 you may find this interesting . 
2 Likes
So this means? It breaks down generated image with ones original prompt or just the image?
Example 1 : I prompt an image ( image generated) … then JSONified Prompt
Example 2: I take an old image generated… then JSONified Prompt
1 Like
The current version of DALL-E does not know how to work with leather. Especially with faces. In comparison with the results of half a year ago.
4 Likes
Did you try to prompt it instead? If the model can’t do it the way you imagined (and I guess this is a problem we all face with DALL-E for quite some things. :-), and thus we may wrongly assume it can’t do it all)
Let me give you one example and then get back to you:
For example I tried to make a portrait of a woman. I saw this as a challenge one day (some weeks ago) when people said: Nuuuh, you can’t do photorealistic pics with DALL-E 3, really. Back then I agreed.
But then I said to myself: What if we just try to prompt it better? find out how it should/could be prompted?
For example the shoes of the woman were not reflecting on the marble floor. So, I just put this into the prompt. And voilá. Suddenly it was possible.
So, back to you:
Did you try to prompt what the leather should look like, etc. already?
DALL-E 3 can definitely do more than we may think. But we gotta prompt it.
Example:
Very simple prompt: Render a woman with a leather cap.
It looks ok, but not 100% convincing for sure.
Second iteration which is: Asking ChatGPT to output the possible improvements in a textual table, work them in, THEN render again.
Let’s repeat this a couple of times… and we get:
P.S.:
All pics were still created with DALL-E 3.
More infos about how to achieve such things and refine them even to way greater detail also here:
In general:
Here I started to write some own things. Maybe it helps you:
6 Likes
My examples are made the same way on DALL-E, but on an older version. But I think DALL-E was not originally designed for photorealistic images. And the latest version does not cope well with 2D.
2 Likes
