In real life, the number of times a pro is going to take a Firefly-generated image and use it as-is are vanishingly small. You can have happy surprises, something I refer to as "controlled serendipity," but of all the images I've ever generated in Firefly, only two were good enough to stand on their own. (Photography is the same way: You start with an idea, and you get as close as you can at the time of capture, but even in the studio there's almost always some adjustment in post-production, even if it's as simple as cropping or adding a vignette.)
Straight text-to-image is so limited that it's almost never fun. Either it takes fifteen tries before Firefly comes up with something cool, or there's an image that's this close but the exposure is off, or there are some of those weird artifacts we know so well, or the composition sucks. You're at the mercy of whatever mood Firefly is in today (and it does have moods, let's face it).
"Challenge" is a bit of a misnomer. There's no challenge in writing a prompt and picking the best of a half-dozen runs, which you would never do in a real (paid) project. You would take one or two or a few into Photoshop and build a composite; adjust exposure, contrast, and color; make some spot adjustments; add elements with GenFill and subtract others. Firefly now has generative fill and other tools to modify a bare result. Why not showcase those?
Here's a super-simple example.
This is what Firefly returned from an initial prompt:
The elements are all there but the parrot is facing the wrong way, ruining the image. (Firefly's sense of composition isn't the greatest.) A few minutes in Photoshop, and we have something much better:
This is why I don't see a good reason to limit these challenges to what comes back from a prompt. Nobody gets a chance to be genuinely creative. Leave the "one and done" images for the gallery on the Firefly home page. A challenge should be something that takes at least a little more work than copy/paste.
1
Reply
AdChoices