hello,

wont try to sound like a complete noob here but I want to go about making a realistic model that'll imitate the newer gen iPhone pics. But first I have some question, prerequisites and input from the community (if they even see this).

First off the reason why I'm doing this is because I'm more impressed with generations that look like REAL photos, like that the average person could take. Not to discredit the countless hour it take to even make a super hyper realistic models, I just have a preference.

I remember seeing the second photo to the Realistic Vision 5 model and made a bunch of images with the tag "Instagram photo"

So, I want the model to be able to do these things:

pink/purple grain effect on iPhone photos

over exposed images

soft shadows

film grain (but closer to apples compression algo)

flash photography

"video screenshot" "tiktok screenshot" "snapchat/Instagram filters" "webcam photo"

(basically different ways to degrade images - not necessarily the UI)

dynamic angles (considering you can move your phone where ever you need to)

I want the model to naturally excel in these (without the use of LoRAs)

selfie hand and arm

accurate phones

nsfw

wide variety emotions and facial expressions

humble backgrounds (ie a simple bedroom with a fan light)

more natural looking people

• I think the people that are generated through AI look fine, but are very same-y and usually (for obvious reasons) look beautiful. I want a model that'll generate a face that you can probably find somewhere not too far from you. I did like what Humans tried to do, but I preferer a more gamble approach when it comes to this rather than trying to generate an image a certain way

Okay, this part is more of a tangent. read it if you'd like but its not necessary. (Skip to questions below this part)

I'm not gonna ignore the fact that there's a lot of ai porn/nudity posed so I will just say with this model I want to be close to amateur/homemade photos you'd find on Twitter or a random "egirls" Instagram who for some reason has thousands of followers.

Questions

1. Is this even possible?

   could a model that's being trained be able to understand things that usually degrades photos (film grain, over exposed, webcam quality)? and without trying to compensate for it? I ask because although that Realistic Vision Image (above) is good and its realistic, its too smooth in places where a normal phone camera would "mess up"

2. I've heard of model merges, how do they work?

   If I fuse a model, lets say Realistic Vision with MajicMIX (for example) how would I be able to make sure the data I want gets kept and the things I don't need are deleted?

3. Is this worth it?

   I know for a fact that id use this but that's not my concern, should I just find/make a grain lyCORIS? (since lyCORIS would probably be better at detecting that I'm tryna train the grain on images)

IMPORTANT: Later on I will put images that I can round up that show what I'm looking for then put them here if Civitai lets me. But if you want to help either, provide images to train, explain a bit more on how to do certain aspects, or have resources that come close to what I'm looking for, contact me here [email protected]