VeryBadImageNegative

Let me be the first to agree, and further add, I grow weary of "mixes, mashups, carefully combined others works, put crap together no idea what's it, etc..."

I did a fully me trained model and it was only 12% useful, so I truly get how hard it can/could/is to make NEW actually NEW trained models with no "mixes added".

And another thing, when people post their "models" and they are festooned (look it up) with loras, it makes my flesh crawl.

/rant done :) (smiley face here).

@GG_System 
https://civitai.com/models/57319/a-zovya-photoreal

https://civitai.com/models/28059/icbinp-i-cant-believe-its-not-photography

https://civitai.com/models/25494/beautiful-realistic-asians

https://civitai.com/models/30240/toonyou

here's four just grabbing from the 'most downloaded' section real quick, should I keep going?

Even then, the ones that had them had like one or two.

Point is, your images aren't showing what your model can do. It's showing what it can do along with a dozen of other people's embeddings and loras and messy prompting. And with that many negative embeddings you're blowing past the vector limit meaning they are doing most likely nothing at all.

Your first prompt starts with : ((masterpiece)), ((Intricate Detail)), masterpiece,

So you're already wasting vectors on duplicate words.

The best part is, this isn't even a bad model. I've been doing some testing with zero negative prompt and getting decent results. Which is how it should be, and how you should be 'meticulously tinkering' your model to perform.

Again, it's not a showcase of the best images you can make using all of your tools available, it's a showcase for THIS tool. Add more images to the gallery at the bottom using all of your resources, but the images uploaded along with the file should demonstrate what the file does by itself, otherwise we're all just left to try and figure out what this really is without your preferred methods of prompting.

@GG_System also worth noting, your uploads aren't really getting downloaded by the TOP CREATORS on Civitai, so @pursuit_of_beauty 's rant is your target audience. Another reason why your example images should be simple and easy to recreate so that your users are able to get started without having to download a shit ton of embeddings and loras.

So many people on here think they're making 'fine tuned merges' when really they're just uploading the same merge and using different prompts and 14 embeddings.

@GG_System just dropped some results to prove my point. It's not that the model is bad, or anything else. Just that you should be showcasing your model itself and what it's capable of on it's own and bringing to the table. It's easiest to do that if you force yourself to remove all of the extra resources, and it REALLY helps with finetuning your model so that you can see how your models react to changes on their own.

Maybe keep in mind that being a creator, you're uploading models to be used by mostly people that aren't creating their own resources. Therefore the level of understanding LORAs and Embeddings varies, and if you want your model to be reviewed well it helps to give people a basic place to start. Get the best images you possibly can using only your model and give those to people to start with, save the embeddings and LORAs for your own portfolio and upload them in the gallery below for more advanced users if you want, but they're just not a good representation of your resource on it's own.

@Fenn  I was perhaps too honest in my post.

My statements about the need for pics to be easy to reproduce (without) a lot of loras, remains correct.

I don't need to have done a bunch of "mixes" in order to be able to state objective reality, (surprise?).

All you people whining should ask for a refund. Oh wait...

@pursuit_of_beauty  It took me 5 itterations and over 100k images, custom tagged and currated, to get my fully trained model to output the level I wanted. It's a pain for sure.