As an Amazon Associate we earn from qualifying purchases.
Anyone else getting incredibly frustrated with the limitations of style synthesis algorithms lately? It feels like we're constantly bumping up against the same walls. I'm working with image generation, trying to recreate the style of early 20th-century photography.The problem isn’t generating old images; it's capturing the specific texture and slightly surreal feel of the orthochromatic film stock from that era.
The algorithms I’ve tried tend to either over-stylize, producing something that looks like a generic filter, or under-stylize, barely changing the original image at all. It's like trying to explain the difference between impressionism and post-impressionism to someone who onyl understands "painting." We're still so far away from reliably achieving nuances in style.
I'm curious if others are having similar experiences in different areas, like text generation (imitating a specific author’s voice) or music composition (replicating the style of a particular composer). What specific limitations are you encountering, and what workarounds have you found, if any? Maybe sharing our struggles will help us all find better solutions (or at least commiserate!).