She looks like she was a screen siren in her younger days.
Freckles are looking a lot better. I wonder how I could train the AI to make port-wine stain style birthmarks, though.
Do you use any modifiers to indicate skin texture?
The problem Iâm seeing is that if I do photorealistic portraits, everything looks photoshopped. I tried to use negative keywords for âairbrushedâ or âretouchedâ and positive for âskin textureâ but it still looks too even.
I want⌠when I have time (ha-ha!) use the AI to rewrite my Space Spinster into a lighter novel that has a lot of weird âspaceâ imagery. For it, Dream-E is great because some filters turn stuff into⌠really weird stuff. So, I think I should be able to have lots of fun.
Here is a portrait I made for my character banner:
Oh, I like that one
Iâve found that getting the desired skin textures (and model - camera distances) tends to rely on luck. Sometimes I can get near-photo quality character portraits, other times theyâre basic 2D paintings (despite the photographic prompts urging otherwise). Maybe the descriptive / scene based prompts also have an effect (confusing or overloading the software).
Fiction will never beat reality ~ unknown quote.
Oh, yesterday I had a run of about ten image results filled with micro-stock style watermarks or business / brand logos. The text was still gibberish, but none of it was asked for. Tinkering with the character description seemed to fix the issue although it was still annoying.
Also, in one request the phrase âmale bodyguardâ was read by the software as âmale bodyâ and flagged as a banned term / prompt. Nice to know MJ considers my physical form to be outlawed. MJâs randomness and inconsistencies make Daz-3D more appealing by the day (at least for character creation)âŚIf only Daz-3Dâs rendering software was AMD / Radeon graphics card compatibleâŚ
MIdjourney prompts for Naida in a forest (see above)
Young Siberian woman, 19 years of age, ginger, orange freckles, strong, muscular, attractive, beautiful, deep blue eyes, youthful complexion, smiling, friendly, green sweat top, standing in a winter forest, portrait photography, DSLR, shot on a Nikon, 24 mega pixels, UHD, ::5 blue eyes
Naida in a forest, head and shoulders shot
^watermark returneth
Prompt: **Full body photograph, Young Siberian woman, 19 years of age, ginger, orange freckles, strong, muscular, attractive, beautiful, deep blue eyes, youthful complexion, smiling, friendly, green jumper, standing in a winter forest, portrait photography, DSLR, shot on a Nikon, 85mm lens, 24 mega pixels, UHD, ::5 blue eyes, full body
Note: **Full body doesnât always work.
Update: If you have access to MJâs discord channel, browse through the image theme gen - character thread. You may find more prompt ideas there.
Oh, I hate the censorship because it doesnât tell you which word triggered it.
On NC, Iâve recently had that problem and apparently it was the combination of âyoung boy.â
How else am I supposed to say it? I want to create a boy thatâs young. If that was paired with something dirty, I can see why, but it wasnât. I think my prompt was something like âyoung boy petting a black kitten.â Just why?
And it took me forever to figure out what the trigger was.
After I finally found the trigger and cut out âyoung,â it still didnât render a boy but a girl or just a cat (or two cats).
I gave up on that prompt and just created a cat instead.
(It was for the daily challenge. I submitted the cat image above and got on a 10% list.)
Today⌠tonight I was on a quest to create realistic skin.
Very long story short, my conclusion is that NightCafe does something to Stable that changes results. I donât know if itâs a filter that they donât admit to or if itâs a problem of computing power but the results of photo-realistic portraits are very far from the raw Stable which you can find for free on Hugging Face.
My final best result on Hugging Face is this. I have succeeded at last!
prompt
prompt: soft focus portrait of a beautiful woman, highly detailed skin texture, chestnut brown hair wavy, thoughtful, mother, forty-year-old mom, tack sharp, sunset in a garden, photojournalism, hazel eyes, bokeh, natural, gentle soul
negative: watermark, cut off, out of frame, monochrome, b&w, airbrushed, retouched, flawless, black and white, painting, drawing, grandma
Meanwhile, using the same prompt on NightCafe:
More prompt weight
Itâs like a freaking Instagram filter.
I know personally people who would read an article like this and take it at face value.
This isnât a problem with AI (though they could build in some safeguards to not use âin a recent studyâ or anything of the sort). The major problem is in gullibility.
Critical thinking needs to become a subject studied from elementary school.
But donât believe me. Believe the pretty words.
Btw, this post was inspired by my news feed that promoted cheese consumption. I thought, I bet I could get AI to write me an article like that and now Iâm wondering if that article actually was written by AI.
It quotes people, etc, but I donât know if theyâre real or if they actually said those things and even if, are they qualified to make those claims.
AI exposes our weaknesses. They were always there. Itâs only now that we see that all this time we shouldnât have been so gullible.
I must share my entry in the âsilly selfiesâ challenge. If I get on the top 5% again, thatâs 20 credits to make up for my âreal skinâ experiments. Wish me luck.
I wonât tell you what type of a selfie it is. You must click to unveil the surprise.
This is the quality content that AI was created for.
Iâve noticed some subtle improvements in Midjourneyâs art generationâŚThe next update should be interestingâŚ
^still being haunted by watermarks though.
This video could be useful for post-production image editing.
This one though looks like an ad byline though.
Stable released a new version (2.1) that works very well for realistic images but Midjourney doesnât use Stable, right? Or do they?
On Stable 2.1 negative prompts work well so I put âwatermarkâ on all.
My recent favorite creation mania was for Melody. I needed a face for a new character so I looked through unused generations to see if I had an interesting face hiding in there and I found this lucky accident I got on the facemorph site long long time ago.
Sheâs gorgeous and I thought it was about time to get her our of the generic âfacesâ folder where all rejects end up and give her the spotlight she deserves.
The image is poor resolution so I was hoping to evolve her. The first attempts werenât going well though.
It always helps the generations if you can correctly identify ethnicity but I have no idea what ethnicity she could be. I got really curious, so naturally, I ended up in a rabbit hole of research of human phenotypes.
If you click on the world map, you can see the different types of face structures common in different parts of the globe. Itâs pretty fascinating. I also very much agree with the common phenotypes of Polish people (Iâm Polish, btw). I know a bunch of people that would fit the East Europid, Nordid, and Pre Slavic phenotypes that are found in Poland. I think Iâm of the latter.
So, clicking around the map, Iâve come to the conclusion that my gorgeous woman was close to Gracile Mediterranid or Trans Mediterranid, which as you can guess, covers people living around the Mediterranean Sea.
I decided to make her Greek in my story as a result of all this, and I looked up some Greek names and decided that Melody Kostas fits her. (she was initially Lily but that did not fit this image at all)
So in my experiments of evolving this image, I tried adding different countries to see their effect and omg, I made so many gorgeous versions of her. Iâve finally settled on these few as the âofficialâ Melody.
Melody business
Melody fun
Melody casual (there was something weird going on with her hairline in several generations. I think Iâm going to have to remove it in an editor)
Melody glam
All of the eyes are kind of wonky because Wombo has been having âeyeâ issues lately. Itâs gotten better but still far from perfect. Not sure what they did to cause that. More work for the editing stage.
Funny sidenote, all of the images were of her looking over her shoulder because my initial attempt to evolve her started from this.
I was just trying to build a taller picture to evolve so I stretched her out and pasted her face over it.
I guess AI interpreting her second neck as a shoulder is better than giving her a double neck.
That needs to have the link very close to it, so I can hunt down the study. If I see the study, and the article is not wholly slanted away from the mess, then Iâll post it.
But that gets tricky.
Like the experiment where they âturned off a belief in Godâ, all the articles at first were super-excited about how belief is just a flipped switch of the wiring in the brain.
But what that thing did was something far more scary. It turned off the same area of the brain that doesnât communicate well in psychopaths. This indicates that the tendency to believe in a higher power is intertwined with the ability to reason situational ethics (higher morality). Its where you decide the order of operations on laws.
Example:
Theft is wrong, but theft to the starving is situational. This is why after Katrina the hard hit areas like New Orleans didnât prosecute those who robbed grocery stores for food, but did prosecute those who ran off with TVs. It was a lawless moment where people were trying to survive until order could be restored, people needed mercy more than justice.
Anyway, they thought they had managed a âGod is just the wiring of the brainâ, but the real outcome when looking at everything, is that they managed to prove that âGod is a required crutch to moralityâ. It 's why other studies show that just mentioning God improves willpower, having a smiley face around Wal-Mart does restrain some shoplifting, because the crutch that has been mocked since before I was born is a hard-wiring thing.
But you wont see a ton of articles connect thatâpartially because many jump to the conclusion that athesits are psychopathicâŚwhich is a hard no (but you dont want an article that appears to say that). Those with that tendency are just LAWFUL or exist in technicalities. They can function in any ideology just fine. They only really expose themselves in situational ethics, which is a daily life thing, not a religious one or lack thereof.
The longer I look at her, the more she looks like Ariana Grande.
Now that you said it, I canât unsee it. Especially on the last one.
Why oh why does my go-to not understand bodycon dresses but can get lingerie down just fine?!
I like the lace sleeves on those dresses
These are the ones I didnât settle on. It was surprisingly good, save for some minor glitches.
The last one looks like Elvira in workplace approved dress