Text in images is mostly a solved problem with frontier models. At least regarding consistency. Most of the time it’s still obvious and looks strange in natural-looking “photos”.
This is an example generated by Qwen Image as a single image file:
It could be both, an ai “edit” of some image pulled from the interweb with text added manually. But it could also be a genuine drawing that was pulled from from the interweb as it is now and the text put over it after. I dont think the text would be it’s biggest challenge, it would more likely struggle with accurately putting the mountains, rivers and borders in the right place without some user provided reference pic. But i can’t imagine the map was drawn for this specific meme.
There’s no they, this is an AI generated image
Idk, image models are usually pretty shit at text and I don’t see any weird amalgamation of h and f that you sometimes see.
Maybe the models are getting better though 🤷
Text in images is mostly a solved problem with frontier models. At least regarding consistency. Most of the time it’s still obvious and looks strange in natural-looking “photos”.
This is an example generated by Qwen Image as a single image file:
However texts are now telltale signs of genAI, due to the fonts becoming emblematic of certain models and the noise patterns on them.
It could be both, an ai “edit” of some image pulled from the interweb with text added manually. But it could also be a genuine drawing that was pulled from from the interweb as it is now and the text put over it after. I dont think the text would be it’s biggest challenge, it would more likely struggle with accurately putting the mountains, rivers and borders in the right place without some user provided reference pic. But i can’t imagine the map was drawn for this specific meme.