One of the things I envisioned when I started the development of this Visual Novel, is that technology would change. Act III already had a major change as models with the capability to adhere to complex prompts were available. An image such as this one would have required a lot of work with SDXL, and with Flux it was effortless.

A different challenge have been the recurring characters, as I didn't perceive a better solution to change the expressions and to recreate the same characters in different clothes or in different settings. I had SDXL loras for characters and I stuck to that in Act III.

Now there might be an alternative, which is Flux Kontext, which can do edits in expressions far better than SDXL taking an image as a base. I think I am going to add this to the mix of models I use. See some results. Base image generated with Chroma:

Without background (Inspyrenet Rembg node) and with closed eyes and open mouth with Flux Kontext.

Change of expression, and a slight change in posture and framing that I didn't ask for, but still works, in Kontext.

An improvement for sure!