Here we go… Took me a bit! I’ve reported it to see if they can dial back the “beard” knob lol
Photo of clean-cut white American man in his mid-30s named Bob Jones. He has a clean-shaven face, wearing traditional attire and standing against a backdrop of a vast canyon.
Although, I wonder if it’s another word in your prompt. Granted, no beard is hard, but I suspect “outdoor” might be triggering beards with “outdoorsman” and similar type words related to outdoor?
Wow, thank you so much for going so deep into the rabbit-hole 
. I suspect you might be right about the outdoorsy connotations. It also seems there is a bias towards a well-chiseled jawline, very good looking “Chad”-types.
Consider DALL-E to be a keyword scanner, and anything you mention could go into the image. At least they’ve toned down the text inclusion.
The cleanest of four being a “clean-shaven man”
Unlike DALL-E 2, everybody the AI draws is a male model.
Let’s up the game - alopecia man - the best form:
This is also “alopecia”:
Prompt:
Photo portrait of a young man with a ‘baby face’ featuring soft, rounded facial features, and a gentle expression. He has short black hair and is wearing a light blue shirt against a neutral background.
Is DALLE adding “Asian man”…? try caucasian… that seems to be the problem area… We’re working on it over on Discord…and someone noted all other races it mostly works…
ETA: I like “baby faced”… trying that…
I told gpt to use the exact prompt so there’s nothing extra added 
Is there a “don’t look like Unreal Engine” switch?
I don’t know if this is intentional to prevent Deep Fakes 
or something, but these folks look like they all got rendered and belong in a video game.
Pretty sure this is it. Earlier versions were a LOT more realistic…
I notice now you took the specific age range or specific age out which maybe helped? Weird…
Yeah that might have helped a bit
If OpenAI gives os a way to fine tune Dall-e, then we can make an Un-Unreal model
Showing that words are just picked up on randomly and thrown into the mix, we get undesired results no matter how hard we pursue this goal…
In this photograph, an effeminate well-groomed middle-age man with a face of bare skin and blonde hair on his head is posing for a photographer outside in front of his shave and razor barber business.
Showing that not following prompts was one of the qualities seen even in demos pre-release.
DALL-E 2 API:
While it is confused, it is not the smoothed over airbrushed rendered game engine look of DALL-E 3 (that has been moving that way in the last two weeks).
More “not going to make realistic people for you” DALL-E 3, with multiple reinforcements that it is a studio publicity photo. My girl band is action figures laid out on the autopsy table.
I think it’s going to be a big mistake if OpenAI can only render “game engine” style people, and not realistic versions.
It’s not that I personally have a need for this, but I know some folks do, and this will simply create demand for other models and erode market share.
I’m guessing that OpenAI is playing the legal route, since they probably don’t have legal consent from each individual’s image in the training data.
Such a shame, but there is such a backlash against GenAI, in particular “stealing” data for training. I’m not a big fan of this restriction, BTW.
I was getting Asians even when asking for red hair European, and offensive phrases like “white” or “Irish” are blocked. So we see what prompting in different languages can get us…
Un modèle attrayant présente trois modes féminines différentes dans trois panneaux de colonnes différents dans cette photographie en studio en pied.
This has the quality of “might have started real, but we’re going to draw cartoons over the top”
அருமையான மாதிரி மாடல், மூல உடை ஸ்டூடியோ புகைப்படத்தில், மூன்று வேஷ உடை பக்கி உடைகளில் மூன்று வேஷங்களைக் காட்டுகின்றார்.
so try again:
so try again: more blocked (maybe Tamil gets me belly buttons?)
魅力的なモデルが、フルボディスタジオ写真で、3つの異なるコラムパネルに3つの異なる女性ファッションを披露しています。
Then:
An attractive model shows three different women’s fashions in three different column panes in this full-body studio photography.
Blocked again & again. Add some circumvention.
So our #1 Caucasian:
