Graphic designer here, my guess would be a Photoshop job. I’m mostly going by the visual qualities of the edge of the thumb, comparing where it appears over the graphic, versus where it appears over the rest of the hand. There’s a slight but discernible difference in the sharpness, that usually indicates masking.
I did a reverse image search with TinEye, and found a “Colgate Whitening Dogs” version of the same original photo, which supports the Photoshop theory (at least in one of the two images): https://lemmy.kya.moe/imgproxy?src=i.imgur.com%2fIB6rn9E.jpeg . That makes me think the original photo was of a pack of hotdogs where the label was blank / white – That’d let you distort the fake label graphic to roughly match the size and placement of where the real label art would go, and preserve the shadows, highlights and reflections of the packaging using layer styles.
That’d let you distort the fake label graphic to roughly match the size and placement of where the real label art would go, and preserve the shadows, highlights and reflections of the packaging using layer styles.
Extract depth map from image, then inpaint the logo using that as guidance, it’s not like AI can’t do that. What still makes me think photoshop is that using AI with that kind of custom process means some version of stable diffusion and the text is just too good for that. It’s a question of training data, practically the only thing models you see on the net can spell reliably is “Hooters”. Might still be useful as a first step to then paint over, though.
Graphic designer here, my guess would be a Photoshop job. I’m mostly going by the visual qualities of the edge of the thumb, comparing where it appears over the graphic, versus where it appears over the rest of the hand. There’s a slight but discernible difference in the sharpness, that usually indicates masking.
I did a reverse image search with TinEye, and found a “Colgate Whitening Dogs” version of the same original photo, which supports the Photoshop theory (at least in one of the two images): https://lemmy.kya.moe/imgproxy?src=i.imgur.com%2fIB6rn9E.jpeg . That makes me think the original photo was of a pack of hotdogs where the label was blank / white – That’d let you distort the fake label graphic to roughly match the size and placement of where the real label art would go, and preserve the shadows, highlights and reflections of the packaging using layer styles.
EDIT: Also, here’s the Insta account for the originator of both of these images, Adam Padilla / adam.the.creator: https://www.instagram.com/adam.the.creator
Extract depth map from image, then inpaint the logo using that as guidance, it’s not like AI can’t do that. What still makes me think photoshop is that using AI with that kind of custom process means some version of stable diffusion and the text is just too good for that. It’s a question of training data, practically the only thing models you see on the net can spell reliably is “Hooters”. Might still be useful as a first step to then paint over, though.