>>8039518Not an expert, but early generative AIs and I think all the currently popular ones were trained on tagged images (and many only 512x512 IIRC), so they didn't understand space and were infamously bad at e.g. hands, helicopters. Now there are multimodal models that incorporate not only text and images, but 3D meshes, so those problems will continue to disappear. Don't know how long it'll take thoughever
>>8039669ten years ago you didn't know these things existed lol