@enzino Without a doubt. I must say, I really admire how you combined the cinematic shades from v4 for definition in the face, and the bright studio lighting from v1 on the body. People want "imperfections" mastered (we all do, creators included) but they don't realize that imperfections also lead to mutations. It's like telling ChatGPT to type like a human; it will end up sounding talking with profanity before it removes jargon. No one defined what "human" or "AI" is, unlike ZeroGPT. THAT is where LoRAs come in. Someone trained something on what defines an imperfection. That defined imperfection is then used as an extension of the base model. The main issue I am starting to see in the community is that a lot of people treat LoRAs like a model dependency package vs what they actually are, which is a Low-Rank Adaptation.

I'mnot saying it's not possible, but who have had imperfections in models successfully it takes paragraphs of succinct prompting tags (which is a total burden if creating images briefly). You'll also notice that if you give them simple sentences, it's often mutations on the bodies and it takes a lot of negative prompts to define it's tag parameters (like Juggernaut Ragnarok).

I am sure that you know most of this, but I also say it to educate others so they know. I have a background in programming with AI (not vibe coding, Python) so I know how they think. LLMs like ChatGPT have people underestimate what it's like to use frontend AI vs create with it. Whenever you do a hyperdetailed model, try to have "golden hour lighting" in the training at a minimal. As realistic as it looks with the oily glimmer highlighting skin detail, it also brings high saturation which overcooks a lot of other models (sadly, like Chloride v2). What you're doing right now using shading to bolden the details is what makes yours better than most.

Nice job on the Amateur model, btw, it's never easy having edges that aren't sharp AND no body mutations. Wicked 1st model release.

@freestuffpl0x42069 Greetings from a tropical island! Honestly, you just convinced me to go ahead with this Checkpoint. Your detailed analysis is what I was looking for and many thanks! I keep one Mojito for you :)

@yann_islander I'm glad! I write these books of comments with good intentions. As you can tell, the creator, @enzino is really good to feedback, which is why I believe their models have a large difference each release. Usually creators will hide the negative comments which makes them look bad, and wastes everyone's time once people use the models and find out flaws regardless.

You should definitely give the amateur model a try. Personally, I like that one a lot due to the analog style which I dig. Don't be disappointed with hyperrealism when it comes to pores; I think that is more of a SDXL issue vs creator training issue. It tends to make bumps or freckles before making pores and that's with every SDXL model that I have. I'm sure pony models can if merged with SDXL, but the tags and weights can be exhausting. :)

all you need is the seed to make it. otherwise, the seed and picture will always be random. If you type anything it will always be random. With this being said, I'm surprised that it looked anything like the initial picture to begin with

@freestuffpl0x42069 Agreed, and I also think the machine (gpu) itself induces randomness.

@enzino you honestly got hands down pretty decent this most recent update of v6. That's true, though. I only know of 2 models that have it down. I have yet to see an SDXL model with pores. It always wants to give a gritty texture which makes it look like sand paper.

@camarcuson194 no doubt. Something fun I like to do is get a few model versions, use the same seed, see how each one differs. It's cool because it often has the same image, but different lighting, postures, textures, or anatomy. That's the best way to find your favorite model. People always prefer one version over others whether it's old or new. My 2 favorites are Pikon Realism v2, enzinos Into Realism Ultra Amateur v1, and definitely this v6. Getting down prompt adherence with natural anomalies like freckles, skin flaws, and the grainy amateur style is always art that stood out to me. You might like those too if you liked the hitchhiking pic.

Join enzino's discord server - the community can help with this for sure

@enzino Yes I do as always.
Is there a specific one you want me to use?, or is it incompatible with your model.

Thanks again.

@the__Realist Hey, I use Forge also. I'm not totally sure what it might be but if it's hyperrealism you're looking for there's a lot of SDXL styler extensions that can get you what you are looking for. Sometimes what will happen is if a skin texture is used multiple times it's overly sensitive. It's the same way for me with "dripping" making drips running down the bodies like an LSD trip watching the walls melt even if I say to have it dripping from a specific body part. Our workflow features are more basic than ComfyUI since they may have node packages that Forge doesn't have.

What I do is I have ComfyUI and Forge on Pinokio so paths are easier to manage and packages are easier to install. Then, I'll use ComfyUI for analyzing workflows after dragging the json to it. If you had that anomaly happen with that image, then it likely happens with others that you generate, also. Thanks for the heads up, I probably make the same sacrifice being on Forge.

Make sure that you're using at least CLIP of 2. I get a black screen on 1.

@chemist2739 That was it! I updated the clip from (-1) to (-2) and there it was! Thanks, man.
I'm not even sure what that is, really.

@enzino Thanks for your reply! chemist2739's response below was the fix for me :) If it happens to others or if it's an issue you see relevant please add it to the description. Thank you for your work!

@tbane526 Awesome, glad to have helped!