So these days I have been getting a kick out of asking DALL-E3 for photos labeled with textual content. They’re simply adequate to be legible, however but:

The meals that will get duplicated appears to range from unfold to unfold.

I additionally requested DALL-E 3 to do the dessert desk.

After I requested ChatGPT-4 (which was sending the picture queries to DALL-E3) what the labels within the above picture say, it answered:

Pumpkin Pie
Pecan Pie
Apple Crumble
Cookies
Trifle
Cheesecake
Brownies

I requested it if the labels had been accurately spelled within the picture, and it mentioned that it could not inform me, that it might solely learn the textual content it gave as a immediate, not the textual content within the picture itself. (That is most likely the proper reply). However once I requested once more, it could say that “Sure, the labels on the desserts within the picture are spelled accurately. Every dessert is labeled with its acceptable identify, guaranteeing readability and correctness within the labeling.”

Bard, when requested to learn the labels within the dessert picture, was not influenced by data of the immediate, and did what I’d take into account a greater job:

The picture exhibits a desk topped with a wide range of desserts, together with:

Pies: Pumpkin Pie, Cockles Pie, Apple Pie, Chess Pie, and Cheese Coales Pie
Cookies: Thinssving, BPOWHIES, BRIINES, and BROESCKE
Brownies: SIEHECADODOVAN and POKY

The label “Posode” just isn’t clear to me, however it might be a reference to the kind of desk or the kind of occasion that the desserts are being served at.

The desserts are all adorned with pumpkins and leaves, suggesting that it is a Thanksgiving-themed unfold.

The picture is probably going taken in a restaurant or bakery, because the desserts are all professionally displayed.

Here’s a devoted studying of all of the labels within the picture:

Posode
Pie
Punpkan Cockes Apple
CREISE Pie

 Pople
 PIE

ChessseCOALES
SIEHECADODOVAN
POKYTHINSSVING
BPOWHIESBRIINES
BROESCKE
DA

This might not be fairly right (and even practically right; I believe it pulled Posode out of the blurry background), but it surely does not less than seize the spirit of the unique. I do like the way in which the bizarre spacing and random code block convey chaos. Perhaps Bard does textual content recognition first after which builds the scene round that, fairly than beginning by recognizing the scene and inferring possible textual content. (Or in ChatGPT’s case, beginning with an assumption about what the picture is meant to include.)

Bonus: A few extra Thanksgiving spreads, and likewise a scene during which your kinfolk are cats.