Claim: AI has trouble generating images of women with salad

Question

AI had trouble recreating pictures of women with salad?

Answer 1

I think there might be something wrong with your AI. From your pictures it seems that AI generated a photo of a woman eating salad, she just looks unsettled. When I gave AI the "woman eating salad" prompt I got some very normal pictures of women eating salad.

https://labs.openai.com/e/SToWksXBUeOXoyru65qUKoK6

Comments

Great answer, as it is a first-person direct test of the claim.

Answer 2

he was able to. however , these pictures were terrifying. it seem like he had no problem making them but them having normal faces seemed a little hard.

https://www.sadanduseless.com/women-laughing-alone-with-salad/

Answer 3

There's no way for anyone to know if what AI creates is really really hard or really really easy. Looking at the image of the woman and the salad it's a person's personal opinion deciding if the image or art is good or bad. Art can be interpreted differently by every single person, therefore it can never be judged as good or bad, troubling or easy. That's a person's personal choice and decision, not a concrete fact. I do see how some images portray women with more teeth or differently shaped hands, but let's not forget that AI is an up-and-coming technology, it can't always be perfect. Humans are complex creatures and it's understandable that they aren't easy subjects to master in a matter of seconds as the AI usually creates images very fast. Overall not all images portray women with more teeth or differently shaped parts, so this can't for sure be said to be extremely troubling.

https://twitter.com/metuiteme/status/1637574988378124289

Comments

I really like our analysis of the claim and explanation. It would be hard to assess whether AI has trouble generating images of women eating salads in an objective manner, especially when everyone has a different opinion on whether the image is considered good for bad. Nice work. The only constructive criticism I would like to add is that if you found an academic paper or research on the topic of good/bad or beauty/ugly and prove that there is no objective criteria to assess that, it would be a nice addition.

Answer 4

I looked up women eating salads on my AI and they generated some pictures perfectly fine. I believe that yours might have an issue where it's not generating any. I also found this website for AI creating women-eating salads. So, overall I would that it is false because it is based on other peoples AI

https://www.creativebloq.com/news/women-salad-ai

Answer 5

I do not think you should be having any trouble having your AI draw a picture of a woman eating salad. Maybe something is wrong? I have had no trouble doing this with a different cite. I will link it below. 

https://simplified.com/ai-image-generator/

Answer 6

https://www.eatliver.com/women-laughing-alone-with-salad/

I found the pictures just fine!

Comments

So how would you rate the claim? True, false, misleading, N/A?

Answer 7

AI does not specifically have trouble generating salads, but it is true that AIs have occasional difficulties with specific human extremities in making them look realistic. The original post that you cited talks specifically about the AI's incapability to make teeth and hands look real, but the salad is also a part of this equation and the question of why the generated image was not realistic.

AI is in its infancy as a technology, so it is understandable that there are some glitches as they are trained with more information and knowledge. AIs are trained with real pictures and databases in order to create more realistic images, and they pull from this information to generate images that otherwise would not exist.

Miles Zimmerman, spokesperson of Stability AI explains in PetaPixel's piece Why AI Image Generators Can’t Get Hands Right that "Part of the reason AI image generators do so badly with hands is that in the datasets used to train the image synthesizers, humans display their extremities less visibly than their faces." He went on to add also the dilemma of positioning within reference images, saying "Hands also tend to be much smaller in the source images, as they are relatively rarely visible in large form." With this line of thought, any picture of a human eating a salad that is fed into AIs presumably feature salads, but not as the focal point. Because it is not as large or detailed as the human faces featured in stock images of people eating salads, AI oftentimes gets confused.

Salads were not the main issue here, but your original claim still initiates curiosity as to why AI generators are better at generating certain things over others. I would rate this claim as misleading as there is more going on here than just salads being a tough point for AI generation.

AI original source: https://pettingzoo.co/@gronk/110070382666359354

PetaPixel source: https://petapixel.com/2023/03/02/why-ai-image-generators-cant-get-hands-right/#:~:text=Why%20is%20AI%20so%20Bad,AI%20spokesperson%20tells%20BuzzFeed%20News.

Comments

So how would you rate the claim?