When iterating through Firefly outputs of similar prompts with different configurations, the system seems to get 'stuck' on outputting the same style (ie. the image is 90% the same in terms of structure, coloring, details, general form, etc. This is despite significantly changing Effects, Color and Tone, Lighting, and Camera Angle settings. I have tested this with multiple reference images with varying degrees of visual intensity, strength matching, etc. with the same results (ie. once a particular style is output, all images start look generally the same). It seems at some point the reference image and prompt configurations are ignored, and it falls into a very 'samey' output style that is difficult to 'escape.' This has been tested across multiple computers and multiple user accounts - this issue seems to be persistent platform wide.
The expected behavior (at least as of a few months ago) is that a switch in even just one of these configuration settings can yield sometimes wildly different results, which was a huge selling point in terms of driving to sometimes very unexpected and very creative places (very, very good thing!). From a technical perspective, I'm not sure what to request back, but it seems like a lot of the chaos / variability (ie. INTEREST) has been removed from the current algorithms and replaced by something much more homogenous.
All of that being said, appreciate all the work on Firefly - thanks all!
- AG