Firefly can't tell the difference between a lettuce and a cabbage.

Report · Jan 25, 2024

I've been trying to produce an oil painting style/still life image of a lettuce, tomato, apple and grapes.
Apparently, Firefly has an issue with this because ALL RESULTS get me a damn cabbage, not a lettuce. Nothing remotely close to a lettuce.
I went to the point of giving an exclusion prompt for "cabbage" and the result simply ignored and lettuce.

Another issue is that at least one of the items in the prompt is always ignored. I either get an apple or a tomato, but rarely I get the two.
Also, the issue of the scales/sizes of the items. I either get gigantic grapes or microscopic tomatos.
I've been refining prompts for hours with the most frustrating results.
I'm not requesting things that do not exist. I'm requesting real things that have very well established sizes and scales and a LETTUCE IS NOT A CABAGGE!
Its infurianting. Completely infurianting, that such a basic prompt is met with such disparity and absurd results.
Its important to state that this is for work. I need the image to be used on a brochure cover and the client request is way too specific and of course, there is no budget to pay a professional oil painter that will take a month to paint. This is what AI should be used for, and all I get is crap results.

Report · Jan 25, 2024

I am not sure if you mean iceberg lettuce or leaf lettuce. Iceberg lettuce and cabbage look similar. I am assuming you mean iceberg.

I took your image and made it a reference image (so it would have the look of your original image) and tried shifting the prompts, but could not get rid of the leaves around it. It still had a "cabbagey" look to me.

https://firefly.adobe.com/public/t2i?id=urn%3Aaaid%3Asc%3AUS%3A22f21127-7efc-49fd-8d4e-8aa2c30bf2f0&...

So, I took the image to Gen Fill. In gen fill I used remove and selected around the head to remove the leaves on the outside of the top and sides. Then I used "insert", selecting the overlapping leaf on the bottom and generated without a prompt to blend it in.

I am not a food expert, so not sure if this still looks too much like a cabbage versus iceberg lettuce, but I think working it in gen fill is likely to get closer to what you want rather than trying prompts in text to image.