Is Stable Diffusion XL Actually a Step Back?

Perhaps it's only been my experience, but after having tried the newly released Clipdrop Stable Diffusion XL 1.0 extensively for the past several days I've found myself sorely disappointed.

One of my favorite methods of image prompting has been to simply to enter a few lines of poetry and then sit back and see what a given AI model makes of them.  As an example, I've in the past had good luck using the opening from John Keats's famous Romantic poem Ode to a Nightingale that reads as follows:

My heart aches, and a drowsy numbness pains

         My sense, as though of hemlock I had drunk,

Or emptied some dull opiate to the drains

         One minute past, and Lethe-wards had sunk:

'Tis not through envy of thy happy lot,

         But being too happy in thine happiness,—

                That thou, light-winged Dryad of the trees

                        In some melodious plot

         Of beechen green, and shadows numberless,

                Singest of summer in full-throated ease.

Whether or not the poetry itself is to one's taste, it's definitely given me some truly inventive images, as can be seen in the example at the top of this post that was generated by SDXL 0.9.  And what's notable is that that older version of the app managed to maintain the same dark style through hundreds of image generations.

Unfortunately, I haven't had nearly the same luck when working with SDXL 1.0.  One need only look at the image below to see the problem.  In place of the edgy images generated by 0.9, I've instead gotten bland generic illustrations better suited to a children's storybook, even though I used the exact same prompt and style (digital art) for both the images shown here.  It's very frustrating, and I've had the same experience with a wide variety of prompts, and not only with poetry either.  It seems the 1.0 AI model is simply not as imaginative as the 0.9, strange as that must sound.

I note that one of the options available when using Clipdrop SDXL 1.0 is the choice of version.  One can opt to go back and once again use version 0.9, but only if one subscribes to the paid Pro plan.  Since the cost isn't all that great ($84 per year) I may eventually decide to go that route.  To me, it would definitely be worth the money if I could once again generate the images I find most appealing.  There are certain other benefits to the Pro plan as well, such as the elimination of watermarks and faster image generation, that make it worth considering.