It normalizes using corporate-owned, corporate-trained, tools that inherently favor corporate perspectives and propaganda.
Even an attempt to subvert capitalist messaging can get screwed up by letting the corpo clanker choose what to draw. The images are just inherently biased and weird.
Here, somehow both panels have a white-supremacist and colonialist undertone: we get a white-savior businessman explaining things to a crowd of grateful Asian people; then he’s standing alone in his formal suit, framed as the only rational thinker in an exotic desert.
I understand that both of these interpretations are perhaps a bit of a stretch, but I truly believe that if OP had taken the effort to design and draw their own imagery for these panels, the imagery would not have had these weird colonialist implications that undermine their intended message.
I do understand that interpretation, although this is not just a feature of AI generated memes. Most memes posted here use only white characters or have western opinionated bias although I don’t see anyone having issues with that, it’s only when it’s AI generated.
The person who created this meme had the option to make the characters in the meme to of various ethnicities but didn’t. It’s not particularly difficult to prompt AI’s in specific ways so it reveals your creative vision rather than allowing the AI to make the decisions for you.
Z-Image is literally made by Alibaba, c’mon now. While you do get access to the weights, and you can tweak it to your liking, you don’t get access to the bulk of the training data that was used to train it, and the corporations are still making the foundation model.
I’ll concede that your workflow is definitely way better than booting up a closed-weight model running remotely via API though.
It normalizes using corporate-owned, corporate-trained, tools that inherently favor corporate perspectives and propaganda. Even an attempt to subvert capitalist messaging can get screwed up by letting the corpo clanker choose what to draw. The images are just inherently biased and weird.
Here, somehow both panels have a white-supremacist and colonialist undertone: we get a white-savior businessman explaining things to a crowd of grateful Asian people; then he’s standing alone in his formal suit, framed as the only rational thinker in an exotic desert.
I understand that both of these interpretations are perhaps a bit of a stretch, but I truly believe that if OP had taken the effort to design and draw their own imagery for these panels, the imagery would not have had these weird colonialist implications that undermine their intended message.
I do understand that interpretation, although this is not just a feature of AI generated memes. Most memes posted here use only white characters or have western opinionated bias although I don’t see anyone having issues with that, it’s only when it’s AI generated.
The person who created this meme had the option to make the characters in the meme to of various ethnicities but didn’t. It’s not particularly difficult to prompt AI’s in specific ways so it reveals your creative vision rather than allowing the AI to make the decisions for you.
Yeah, with a careful prompt it can work, I guess. But the assumptions it makes around your prompt are insidious and sneaky.
I saw it as two different people.
His hair and face seems different, with a darker tone.
I see it as Westernised non-Westerner, noticing the lack of attention and support.
please explain how running an open model like Z-Image on my machine normalizes corporate-owned tools, I’ll wait https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
Z-Image is literally made by Alibaba, c’mon now. While you do get access to the weights, and you can tweak it to your liking, you don’t get access to the bulk of the training data that was used to train it, and the corporations are still making the foundation model.
I’ll concede that your workflow is definitely way better than booting up a closed-weight model running remotely via API though.
Ok it’s made by Alibaba, you don’t get access to original training data and how does any of that actually matter?