I'm not sure if this qualifies as a bug, but I entered "Teddy bear with balloons and text Congratulations!" in the GenAI engine and got the result you see.
Ahem 😅 It is known that Firefly isn't yet very good with actual text gneration in this sort of context. Generating the image of the bear and balloons, then adding text in Adobe Express (other programs are available e.g. Photoshop 😉) is a more reliable way to get the right sentiment across.