Sorry, as I now mostly use Fooocus, the metadata are not saved within the PNGs. They claim this as feature :-( But it saves some data in a log.index. I searched this and added as comments. But my experience says, it is hard to reproduce the exact same images with different tools like automatic1111 or even different versions of Fooocus. My guess this comes from several factors:

Fooocus has a lot of styles that are on per default and use GPT-2 to extent the prompt somehow.

It also hides sampler parameters etc. and maybe useses seed and sampler in different ways than automatic1111.

But the prompts maybe give you some clue. They are in no way very smart or elaborate. I also think that Standspurfahrer reached much better quality with my LoRa, than what I got out of it.

This was my very first try to get a LoRa and I didn't go to deep into how to do it right. I just throw hundreds of images at it, that I did over the last weeks with Dall-E3. I really like the creativity and style that DallE3 give with my weird prompts and my wish to have a somehow sketchy style (for photorealism I prefer SDXL and Fooocus much more).

I tried to replicate this in Fooocus with mix of inbuild styles and LoRas, but never got it right. So the idea was just to bake the Dall-e3 style into an SDXL LoRa. But maybe I need to invest more time in selecting the right images (I mixed up several illustration and sketch styles) and also add more description to what the images show (although my thinking is, that style should be separable from content).

I'm open to suggestions how to improve the LoRa making process. If you know some good guides, how to do it in Civitai, I would be thankful to know.

I guess my own GPU is to low on VRAM to create LoRas locally - but even this might be wrong.