I know SDXL is the future, and I will transition to it eventually. But at this moment in time, these well trained models performs far superior for my uses.
✨ Evolution Perfected: The different versions are all about fine-tuning a model that's both user-friendly and sophisticated. There is no need for countless positive and negative statements. Yet more intricate prompts are welcomed as well.
🎨 Mastery Over All Domains: Whether it's the enchanting realms of fantasy or the grounded details of realism, whether it's the safe haven of SFW or the daring edges of NSFW, it has got you covered.
🌍 A Global Perspective with a Local Touch: The model isn't just a jack-of-all-trades; it's a master of specificity. Seeking the essence of Japanese, Vietnamese, or Thai aesthetics? It's got it. Looking for the unique vibes of German, Icelandic, or American? It's all here. From the fiery passion of Italian to the vibrant beats of Brazilian and the rich history of Mexican, the model celebrates the diverse tapestry of our world. And if you're thinking broader, it still handle generics with the same finesse.
Unveil the magic. Experience the power of understanding both the specific and the generic with Unreal Realism.
Now I can hardly take credit for V1, that was simply a merger of:
These are the giants, on whose shoulders I stand.
Description
Easier than ever, to get high quality realistic results.
FAQ
Comments (11)
Other than the initial merge, what was done for the updates? Is it just changing ratios or adding models, or is there some actual photo training going into it? Looks great and you really picked a few of the very best to merge.
I am still rather inexperienced with actual checkpoints, but what I have learned, is how to train LoRas, by following brilliant guides, both here and on youtube. And when I have successfully trained LoRas, to "understand" aesthetics I like, I have merged them into my main model, sometimes it results in the model producing better results, sometimes it doesn't. I am still learning/experimenting. But every now and then, when I think the model provides better results consistently, I share it here, as a new version.
So it still originates as a Frankenstein/merge of other peoples work, as specified in the description, V2 was a re-merger with slightly different ratios. Since then, my own LoRas have been added.
With little luck, I have also put quite some effort into trying to get the model less reliant on negative prompting - however I find that very hard to do. It seems no matter what, I get best results, with negatives. Even if what the negative prompts describes never shows up in the output, somehow the model responds better to the positive prompt, when the negatives are there.
Any guides/articles on the matter would be most welcome.
@Repsirch Thanks for the info! I have integrated some info from an XL article into my own ultimate negative, but I really think 1.5 needs them more than XL. The article is here: Civitai | Good universal negative prompt for SDXL 1.0
I have tuned those for 1.5 and many seem to help my standard negative without much downside. Keep an eye on some of the top new releases, and see if you can reach out to the creators, they may have some insight. Many of them say they don't need much of a negative.
Are you going to be adding a smaller pruned version of V4 model?
It's added now
Are your shure it support German? I did't see sandals and socks. Great Work I will test it.
Maybe someone has encountered... If you generate an image and a person is depicted in full height with a face, some kind of crap starts to happen. Tried a bunch of settings, but nothing changes...
I have never seen this, so it would be really helpful, if you could provide a prompt where this happens, so that I can look into fixing it.
@Repsirch Prompt
Blonde Instagram photo, legs
Negative prompt
cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
Sampler
DPM2MKarras
Steps
39
Cfg scale
7
Width
512
Height
768
Seed
Clip skip
1
Base model
SD1
v1,2,3,and 4 are all excellent checkpoints
very great model thanks
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.