My model is a merge of Illustrious checkpoints. The Illustrious model, along with Pony and Noobai, all belong to a new "family" of models.

​Family 1: Standard SDXL 1.0 (e.g., Base SDXL, Juggernaut, RealVis)

​Uses descriptive, natural language prompts.

​Family 2: Pony/Illustrious/Noobai (e.g., Pony Diffusion V6, Illustrious, Noobai)

​These are not simple fine-tunes. They are heavily modified forks of SDXL.

​They were all trained on a dataset using Danbooru tags.

​This means they require "magic" or "quality" tags like score_9, score_8_up, masterpiece to get high-quality images.

​So, listing it with "Pony tags" is the correct way to use this model. The tags are not op tional.

What Loras did you have trouble with?

Just to make sure I will test various Loras once I get home, but thanks for the feedback, let me know what Loras you had problems with if you're able to, so I can test them as well. 

@Fatbuns The score_9, score_8 tags are part of a node in my workflow, I left them on along with masterpiece and illustrious quality tags (I was unaware of score_x being solely based on Pony)
But I find it odd that you're making the conclusion that the model is pony solely because of the prompt? You have not told me what lora you had problems with? The loras I used to achieve the baked style where all illustrious. The models in the merge are all illustrious, NeroMatura, Waillustrious, amanatsu, and the rest of the iterations are all illustrious, I don't have a single pony checkpoint in my nvme.

@Fatbuns I am not an expert, that's why I am asking, my only point is on you saying this model is pony which is not possible.

The other aspects of your claim I concede that I am wrong but solely on the claim this model is pony in that regard they don't change that.

I am not arguing to be right, like I said I am not an expert I've posted this in various of my models, I've even stated I am not good at making Loras, I am constantly testing and trying things on my workflow. I am always open and constantly reading slowly the things that I need, it's a lot of information and I try to grasp as much as I can, I've only been deep diving on this area of AI for barely 5 months.

Your sheer imperiousness is what annoyed me. As some one who has Neuroticism and is high on agreeableness, I never argue in bad faith, I simply hate arrogant despots who want to belittle for the sake of their own ego.

There's quite a few arrogant and highly entitled people here on civitai, and it's understood to some extent, but getting random DMs and negative feedback with no leeway, getting no feedback to work with just straight up "I am better than you" type of stuff, while being impossible to please the need, it gets tiring.

I am not here to compete with anyone I just share what I like and what I make while I learn in the process. I don't see what you're discussing or arguing about since I already told you the models used for the merge are all illustrious, I don't have a single pony model in my drive, yet you are still stuck on stating this model is pony.

Unless Wai's or the other models are wrongly listed then yes, but none have the pony listing, nor do pony Loras work with them, I've tried a few older pony Loras and they all came out baked or blacked out.

So I don't see what your goal is?

This model is not pony.

@LatteLeopard I only had the 'score_9' tags on the prompt off scene composer node in comfyui, they where just ignored by the model since regenerating without them didn't change anything. I don't know where Fatbuns got that the model was pony only based of a simple prompt tag. Invalid words in a prompt don't affect the model as a lora would.
Even if the embeddings are pony it doesn't matter.

Model Architecture vs. Inputs: My base model 'illustrious' checkpoint. This is the core architecture. Prompts and embeddings are inputs to that architecture. They do not, and cannot, change the model file.

LoRA vs. Embedding: If it was a pony LoRA, which does inject new weights into the U-Net then YES, but I did not use one. An Embedding (Textual Inversion) only manipulates the text encoder's existing weights. It adds no new data.
Prompt Tags: score_9 is a tag from the Pony training dataset. On a non-Pony model, it has zero function.

I was unaware of those tags since I don't even use Pony, so I just left them on the prompt composer when making the back images, all it did was get ignored, since I also had the Masterpiece tag.

@LatteLeopard my first response was directly off gemini, I took my time and read more, so that response was incorrect, what I meant to say was that they are based off stable diffusion, they are different architectures as you stated, I have attempted to merge NoobAI with illustrious in comfyui but it doesn't turn out very well, I realized the reason now xD

@LatteLeopard Interesting enough this model
https://civitai.com/models/1902466?modelVersionId=2153424

Has a Pony lora highly baked in for the overall style I don't know how it worked xD