feedback to the quality of firefly

Hello Adobe Team,

I am a user that from time to time uses firefly to generate images, in often a monthly interval.

Today I got the feeling that the quality of the generation model has decreased.

I have this term adjusted over multiple iterations with firefly v2

"a sports car that gets its engine pulled out like its on repair. The motor of the car should be hanging above the open bonnet on gray crane-cables that reach to a crane on the upper part of the image. The engine bay should be empty and motor and empty engine bay should have similar size. Furthermore the car's interior is dissassembled and a toolbox is visible. futuristic colors, design and rendering. realtistic physics and detailed parts."

90% of the images created result in broken car models, and usually motor-blocks that have the size of another car (when not actually being a real second car - that happened as well), while the bonnets are attached to the roof of the car.

I have now added every edge case and everything I can think of into the prompt.

still the rendering results in something like this:

from which i took the last image since at least the bonnet has a somewhat fitting size and shape, although it is connected to the roof.

Maybe you have some tips for me on how I can create better prompts, or prompts that can be followed more closely by firefly. I know the "suggestions" option in the latest firefly version, however don't find it's suggestions helpful on most of use-cases.

Regards