How to modify AI generated objects, preserving its look?

[maracujaneiro]

When I've generated an object with Firefly's "Text to Picture" AI, and when I'm very satisfied with the result - how can I modify that object preserving the same style or "look"? Or, how can I generate another picture of same or similar overall style, containing the previously generated object? To give an example: When I've made Firefly generating a fantastic animal, I would like to generate very similar pictures, with more animals of the same type for example or with that animal in another pose or with another background.

That way, I could make a picture story, focussing on a main object. (Regenerating a picture with slightly modified text doesn't work.)

I would appreciate any hint.

[droopydog500]

Here is a great hint!  Use the image you like as a style reference for the next image:

[maracujaneiro]

Thanks a lot, droopydog500,

I still tried that option, and so the overall look remained similar, but not similar enough. Especially the main object I was focussing on was changing too much. Is there an option to control how strong the reference image is applied to the new one?

Best regards.

[droopydog500]

Yes, it is just above the Style reference box. It adjusts the strength of all styles applied, including the style reference.

[maracujaneiro]

Thanks a lot, droopydog500:

Indeed, I didn't notice that there is a second option "Strength", because it only appears after selecting or uploading a reference image. With that option selected, the result is pretty good. Here is an example:

Image 2 was created with image 1 as style reference.

What I'm ultimately looking for is an option to apply even more strength to the main object (the rabbit), and less strength to the background. Now I'll try to get some "better" results with Adobe's Generative Fill an Generative Expand. Or do you recommend another workaround?

[droopydog500]

You are basically asking for a feature that many have asked for which is to be able to identify an object like a character and have that object appear in multiple generations in different ways (like poses, etc.). That feature does not exist yet.  The work around I gave above is like the weakest way to do it which has the deficiencies you noted.

[maracujaneiro]

Exactly, I would like "to be able to identify an object like a character and have that object appear in multiple generations in different ways (like poses, etc.)".

Adobe is announcing "Generative Audio and Video":

So I'll stay tuned, hoping that feature will bring me closer to my needs.