Amazing Reality for Krea2

@Liant  I could describe in detail what "your" checkpoint likes and dislikes, but instead, I kept the prompts visible and employed a variety of styles and cues.

It’s great to write with.

@PrincesOfDarkness TBH you descibed a general Krea2 engine's problem in the first phrase. Krea2 is very artistic checkpoint and personally I compare it with Chroma. But for realism sometime you need to try hard and sometimes it's rather difficult to archive something close to photorealism. "Photorealistic" outputs often look like screenshot from modern movie full of computer special effects.

@mphobbit "Photorealistic" people assume this means one thing/the obvious; but it doesn't. Consider the context.

Have you ever heard someone describe an Instagram photo as "Photorealistic"? No. Because it's completely redundant, & would be silly. It doesn't make any sense to say a realistic candid photo is realistic....because it's just a given.

Now who actually uses that word/where is it frequently used....? It's frequently used as a marketing buzzword to describe big budget computer games, with overblown post-processing, etc. There's the problem; that's the data source it's pulling from what you ask for "Photorealistic/photorealism".

I think a lot of people have picked up a ton of bad habits from SDXL, FLUX, etc; "Masterpiece, photorealism", etc; just a load of junk that often does the opposite of what people think it does.

A lot of AI LLMs/Vision models are guilty of this btw, so just beware. When you ask for "Give me a good prompt for Krea 2".....the internal thinking often translates as "What's Krea 2....? It's probably an image checkpoint, probably like SDXL....I'll use a prompt for that instead because I've never heard of Krea 2 (because it came after the data it was trained on), it's probably quite similar". Then you get landed with a junk prompt, wondering why it sucks.

I'd just avoid using any abstract word salad in Krea 2. It does more damage than good most of the time. It's actually a good thing (strong prompt adherence); but it just looks bad because the end result often produces a mess (usually a ton of tags/data all pulling in opposite directions trying to produce a coherent final image output; which is what people asked for, just without knowing it).

@mracar I’m not sure what you’re unable to do—could you describe the problem in more detail? The model is quite demanding when it comes to prompts, but generally, most things work fine without needing LoRAs.

I loved many features of the images created with the model, from the framing to the way they looked like works of art. However, often (much more often than normal with Krea2) stains or beads or liquid form. It was rarely possible to achieve images like your visual examples (with the same settings as yours). Although there is no liquid-containing medium or the word liquid mentioned, I am talking about the liters of white liquid that forms. There would be no problem if the liquid blocker LORA worked, I don't know why but it does not work in this model. Other than that, everything is very nice. Despite everything, thank you for preparing a beautiful model.

@mracar  Reduce the step; do not sharpen details. Show prompts :)

litrally same every model allways has sweat or somthing on the body no matter what

@PastelPastelPastel  https://civitai.red/images/139108663 Do you see any such signs on the images I’m sharing?

I understand. Share your prompt—I'll take a look at it and maybe rewrite it in my own way.

@PrincesOfDarkness its not your model its krea2 in general it just looooooves wet body lol

@PastelPastelPastel I never wrote that it was mine; I just showed you how to write prompts.

@ShadowCell yeah thats what im talking about the covered in water thing the pompts themselfs have nothing about water or goo in them even putting negitive prompts for anything water related dosent get rid of it