Images including people are severely culturally out of touch

[CJWex]

The prompt: 'A person who cannot sleep. Vector illustration.'

Content type: Art

The result:

This is absurd.

I am located in Ireland. I understand that Adobe uses my profile to give what it thinks are culturally appropriate results, but this is way off. Beleive it or not, Irish people are not leprechauns.

Please prioritise cultural awareness and update your system so default results better reflect reality.

[droopydog500]

This issue has been discussed in this thread:

Union Jack appearing in images 

The decision Adobe made is to try to make results appear more culturally relevant based on where the user is located. There are advantages and disadvantages of doing that, and in some cases, the results can be a little ridiculous as you observed. You can change your prompts to be more descriptive to override that region/culture behaviour.

[CJWex]

Thanks very much for your reply - I had seen other similar threads, but not that one.

I understand that you can change the prompt to be more specific (this is not a great prompt anyway), but what I'm arguing here is that the default result is culturally inappropriate. While this example is obviously a long way off, more subtle examples may be equally incorrect but not as obviously so, and there lies a problem.

[Emrah321140800nb5]

I want to create artificial intelligence visuals without country borders. Arab women and men appear in the images I create. (unless I specifically state) There should be no country borders in the images or country selection should be optional

[CMass]

@CJWex Thank you for reaching out and sharing this issue with us. The appropriate team has been made aware of this issue and will get this fixed in an upcoming release. If you see anything similar, please be sure to flag it. 

@droopydog500 thank you for your help. It seems like we are running into two issues. Please feel free to follow this post if we have any updates. 🙂 

[Inush Digital]

I'm always getting Ireland-themed results for some reason, my people have green shirts and hats with club even though I don't write that kind of prompt, and I can't figure out why.

[Rainer3547560048j3]

Hallo Adobe!

Firefly Image 2 macht mir leider Probleme... Egal was in meinem Prompt steht - in fast jedem Bild sind die Alpen oder andere Berge im Hintergrund zu sehen. Auch unter erweiterte Einstellungen als Ausschluss-Kriterium eingegeben. Bringt nichts.

Habe gerade eine Stunde lange mit einem Ihrer Mitarbeiter per Remotedesktop herum probiert, er gab mir dann diesen Link weiter weil er auch nicht mehr weiter wusste.

Das hat's jetzt ungefähr seit einer Woche und es wird immer schlimmer. Beim Prompt "child gives mom a bir red heart, love, mother's day" sitzen die beiden am Berg!

Merkt sich Firefly meine Präferenzen, glaut Firefly von mir, dass ich Berge bevorzuge? Und kann man da manuell eingreifen - meine "Vorlieben" sozusagen rauslöschen?

Bitte um Hilfe!

[rayek.elfin]

It's incredibly biased and in a very VERY wrong way. I hail from Vancouver BC, and "obviously" most landscape images should include 'rocky' mountains with coniferous forests.

There are other issues, of course, and with all stable diffusion bias is problematic.

That said, when I try the prompt

peasant running in a medieval landscape and a majestic city in the background, sun is setting, golden hour

and use default settings this is what Firefly gives me:

Compare with Krita's AI plugin and Cinematic XL diffusion:

(I did have to introduce "woman" as a keyword to create the woman peasant: bias!)

Firefly's result is somewhat laughable. The straw hats make no sense, nor do the modern running shoes 😉 The runners look pasted into the image in Firefly's case. And what's with their white blouses? How do those fit in a medieval setting?

Compared, XL is quite convincing, however. Also biased, but quite usable. Something out of a European film set in a medieval place.

Even with muted colours Firefly can't resist adding in ridiculous aspects:

 

Suddenly modern apartment buildings pop up in the background as well. Runners are still wearing Nikes...

And mostly the results remain unconvincing and plain odd looking. Why is everyone running away from the camera?

Creating a digital painting using the same prompt results in these (FF on the left, XL diffusion in Krita on the right):

Still ways to go. I am aware I am comparing apples with pears, but still...

Free open source seems leagues ahead and is improving rapidly (unfortunately disregarding artists' work, however).