Quality of image output needs improvement

@Gustavo29369441mrq9 Merging all threads here so the team can track where we need improvement

 The AI ​​is pretty cool, fluid and friendly. I report that in the generation of image by text, the AI ​​is having difficulty understanding what vertical stripes are. In fact, it is having difficulty differentiating horizontal stripes from vertical stripes.

@Sean29358192fddx Merging your thread here so the team can access all comments on this in one place.

In trying to generate images of houses, I noticed that the image model doesn't seem to understand different architectural styles. For example, "midecentury modern" will return images of modern box-style homes, but not the classic midcentury style. "Split level house" or "atomic ranch" likewise don't return anything resembling these styles.

I'm currently trying to create an image of a meadow with a sad and depressing atmosphere. The most important part for this prompt is that the meadow is full with flowers that have been stepped on and are therefore broken. I have barely managed to get any pictures where the flowers looked even remotely damaged and the more I tried to improve the prompt the worse the pictures got. There have also been some words that supposedly violate the guidelines , one of which seems to be "trampled"

Here's some prompts that I've tried:

Create an image of a meadow with trampled flowers and a gloomy atmosphere. The picture should convey a feeling of desolation and loneliness, and there should be no people in the scene. The sky should be overcast, and the colors should be muted, with a sense of sadness and emptiness.

Generate an image of a meadow with flowers that have been stepped on, creating a desolate and lonely atmosphere. The picture should have no people present and convey a gloomy vibe. The sky should be overcast, and the colors should be muted to give a sense of sadness and emptiness

Generate an image of a meadow with flowers that have been flattened or damaged, creating a sense of sadness and destruction. The picture should have no people present and convey a gloomy, somber atmosphere. The sky should be overcast, and the colors should be muted to enhance the feeling of emptiness and desolation.

Depict a meadow scene with flowers that appear flattened or damaged, conveying a somber and desolate atmosphere. Ensure that no people are present in the scene, and use muted colors and an overcast sky to enhance the feeling of emptiness and sadness.

I added to this: a line etching of a waterfall leading to a river full of rocks in the foreground with tall fir trees leading into a forest in the background, with mountains obscuring the sky all inside a dogs head. I added "all inside a dogs head" to the end. It didnt show any dogs or animals of anykind.

Whenever I try to generate something with a lot of details the AI will ignore some parts and kind of just do whatever and leave out details making the image look badly photoshop, deformed or prioritize some parts to look better.

Here is the prompt: full body in frame rogue woman in a standing pose wearing black and red leather armour in a forest background, forest background has dark autumn colours with trees, the ground is made with small rocks and dirt

a) 3:4 -No styles applied: https://imgur.com/a/cWuSXYw

It applied the bokeh effect on all of them, 2 images aren't women, some have a darker background like i asked "dark autumn colours" but others have a much more brighter background. The ground and clothes has a lot more detail and shading than the faces and hands.

b) 3:4 -Styles applied: Photo, Vibrant colour, Wide Angle, Studio lighting: https://imgur.com/a/ZTy5Ip8

Ignored the "rogue woman" again, the dark autumn colours are present but maybe because of the studio lighting style, the faces for the first 3 look more realistic but the males body don't look like they fit in the clothing properly so the arms are a bit deformed. The ground is again extremely detailed and life like but the bokeh effect in the background is still present despite me putting wide angle style in it.

c) 1:1-Styles applied: Photo, Vibrant colour, Wide Angle, Studio lighting: https://imgur.com/a/alh7pgv

It's pretty much the same from "b)" but the background is a lot more darker. The weird thing is that the ai made a woman with her breast out but it's blurred??

d) 16:9-Styles applied: Art, Cool Tone, Backlighting, Chaotic, Beautiful: https://imgur.com/a/OnYPZLq

With this one it just went all over the place, it mixed art with photography and body anatomy is very deformed

e) 4:3 -Styles applied: Art, Fantasy, painting, vibrant colour, doodle: https://imgur.com/a/5EfOw89

Styles pretty much ignored, it just did its own thing.

Too long prompt doesn't work well, even Content type or another settng too.

Watch my example for short prompt and long one in same settings.

Fairly works with 2 characters of different aesthetics>

caveman with guitar, dinosaur with drums or any combinations of characters.

If you want more than 2 it won't work>

caveman with guitar, dinosaur with drums, alien with flute

1 caveman with guitar, 1 dinosaur with drums, 1 alien with flute or any combinations of characters.