Making consistend vertical fantasy card border

[crisp parcel]

Good evening,

I am trying to create a fantasy card game inspired from the french MMORPG "Dofus". In this game there are 19 classes (characters). I want to create one card per class.
I finally made it with one class and I'm really happy with the result (image attached). In addition, the card border looks great to me and I'd like to re-use it, or be as close as possible from it, for my other cards.

How can I make sure to keep this consistency for the card border ?

Actual prompt :
Vertical fantasy card, image featuring a character inspired by a Huppermage from Dofus, set in a very old library stands facing away from us, draped in an ornate robe. The mage will be surrounded by a cycle of elements: rock, water, fire, and air, flowing in a continuous loop. The library will have an ancient feel, with rows of old books and possibly tall, arched windows. Infinitesimal etchings on the walls and surfaces of the library will add to the mystical atmosphere, depicting elements and arcane symbols. The artwork is framed by a decorative card border with the text "Huppermage".

[potent whale]

The only way to keep exact consistency in AI txt2img generation is with something called "inpainting." This means keeping some portions of an image literally identical, and only changing an indicated section of an image. Right now, DALL·E 3 on ChatGPT doesn't support inpainting, so the closest you can get is "pretty close."

What I'd recommend is: take this image, and upload it to GPT-4. Ask it to give you a thorough and comprehensive visual description. Then, replace the parts of the description with the new details you want, and ask it to make images based on that description. Alternately, I think you could skip a step by uploading the pic and asking immediately for variations on it based off of your desired descriptions, but you might lose a little bit of control not editing the whole description yourself.

Your mileage may vary on all of this, so it's really just about experimenting at this phase of DALL·E! Also be aware that there's a glitch you might have already seen where asking for a tall image will actually return a wide image that's just been rotated (and mirrored) in a tall orientation, so you might see that as you experiment.