Considering it's their first trained model, and they picked a really tricky base, it would make sense. I'm sure they will get better at it along the way. it's all about trial and error.

@rbbcck381

How is that useful information btw?
Comparing a model that was probably fine tuned with very bland and base model with a model that got merged hundreds of times. lmao

@Part_LoRAs ????? it wasn't me that bashed it

@CrazyGirlAshley why are you all bashing me when im defending you.......

@Part_LoRAs and @CrazyGirlAshley please check and reply to the correct person .it seems you replied to the wrong person i guess the account R logo makes it seem like I'm the one who posted the review. I train and merge models and know how tricky it is, that's why i backed up OP under this review lol

I put a profile picture now so you can differentiate easily

@dark_aromaa sry mb lmao

hey, so, can tell you're new to all of this 'rbbcck381', if that's even your real name.

Can't believe you're over here posting comparison images of a trained model next to a merge and then being a dick about it.
You're basically comparing raw ground beef to a fully cooked burger.

This is what's called a 'TRAINED' model. Most 'trained' models are hot piles of trash on their own. There are plenty on here that claim to be trained models, but most have been through several merges before their creators upload them, diluting the original dataset with more fine-tuned merges. If you see a trained model on here with few generation flaws, that's likely due to it being a merge. But you're straight up comparing with Acorn and that's blatantly a merge.

ALL OF THAT SAID, hot piles of trash still contain new original datasets with information not contained in the thousands of merges on here. And MOST of those merges are very similar to each other due to being slightly different merges of the same base datasets. THEREFORE this model is MUCH more valuable to the community than yet ANOTHER merge. There is data in this model that doesn't exist in any other, and chances are merging it with others would add new elements not seen elsewhere in a much more polished fashion.

So, in short, you're being a dick about things you don't really understand and owe OP an apology. And should delete your useless comparisons.

@Fenn  I have nothing to apologize for. They posted the model and tagged it as a photorealistic model which it just plainly isn't. The whole point of a review is to let people know what they are actually getting if they download a model. If someone clicks the photorealistic tag because they are looking for a photorealistic model, then the comparisons will show them what they will get with the model vs others that are better. If what they want is a raw model that they they intend to merge with others to see what they can make, that is fine too, but at least they will know what they are downloading. If no claim of photorealism had been made by the uploader, I would not have marked it down in my review. I download models because I want to make good images with them, this one is very poor at that, so that is reflected in my review. If it eventually becomes a good model, I will give that version a good review.

@Part_LoRAs The comparison is useful because the model uploader tagged it as photorealistic. If people come here to look for a photoreal model, the comparisons show them the output of this model compared to ones that are better. It can save them time and hard drive space for those who have very limited space. There is no malice in my review, just a statement of fact and comparisons to some models that are, at this time, much better at generating photorealistic images.

My primary interest in AI art is in photoreal images. I really only review models that claim photorealism. The comparisons showed some of the ones I have used before that are very good. The best one's I have used are probably Acorn V2, EpicRealism NaturalSin, Realistic Vision V51, EpicDream Lullaby and a few others. If this one eventually gets good, I will review it positively as well. For now, it's not good.

@rbbcck381 Well, I don't think this discussion is useful at all.
BY DEFINITION, this model IS PHOTOREALISTIC. It doesn't matter at all what you think.
There's nothing to discuss about it.
Model output QUALITY and what was trained with are two different things. If the model is indeed a fine tuned model with 3k images using a base model that got trained with MILLIONs of images, you can't really expect it to output high quality images without merging it.
Base SD 1.5 got trained mostly with 256x256 resolution images to learn concepts.

The point of base models like NAI, Base SD 1.5, SDXL 1.0 and URPM is to have a general and unbiased understanding of literally everything.
The point of fine-tunes is to improve on what those base models already knows about those things.

You don't like the model outputs? Why are you browsing through the CHECKPOINT TRAINED category?
Just filter it out.

@rbbcck381 you're missing the whole point. This isn't a merged model, it's a trained model. This is standard quality output for trained models.

"If no claim of photorealism had been made by the uploader, I would not have marked it down in my review"
Also, missing the mark here. The description makes absolutely no claim to being 'tHe BEsT pHoTOgRaPHy MODeL', if you're referring simply to the keyword tag then you're overreacting.

The description simply states a little training info and that's it. If the data it was trained on are photos, then the keyword tag is accurate.

So yet again, if you're looking for more polished models, then keep your model filter set to 'merged' and quit coming to the raw meat aisle and complaining there's no burger.

@rbbcck381 worth noting, I'm not trying to be a dick, just trying to explain the value of these types of misunderstood resources. Hate seeing quality uploads like this get discouraged because of unfair comparisons.

@Fenn I never said the model has no value, my review only points out that at this time it is not a good model to generate realistic photo images. Period. Nothing else. The comparisons help those people that aren't experts to know which models to download if they want to generate quality images. I, and I think the majority of users of this site, check for the newest models first. If they are looking for a quality photo model, they can see what they would be getting with this model vs some others that might be more what they are looking for. Currently this model produces crap images. If it is better in the future, great. Then it will get a good review.

@rbbcck381 jesus christ I dont know how else to explain this, you're looking at 'newest' first in the wrong category bud if you want "quality images".

Filter it to 'merge' or you're just going to leave a shitty review on every 'trained' model that you still dont understand.

@Fenn I don't need anything explained by you. I leave the reviews for those who don't know. Not everyone is an expert, or thinks they are. If someone is looking for a model that gives good results, this shows them what they actually get with the model. The review is based on the results from the model being reviewed, not for a direct comparison with the better models. The comparison just lets those who might otherwise download this model thinking it is going to give them a quality result, that it likely will not do what it is they are looking for.

Please god, how do I unsub or block notifications from a thread. This is pointless.

I hope to be backup and well received when I finally decide to release my Anime fine tune.
Highly biased datasets composed of 5.6k images with manually tuned tags for every single image.