Preferences

starchild3001 parent
This feels like a real inflection point for image editing models. What stood out to me isn’t just the raw generative quality, but the consistency across edits and the ability to blend multiple references without falling apart. That’s something people have been hacking around with pipelines (Midjourney → Photoshop → Inpainting tool), but seeing it consolidated in one model/API makes workflows dramatically simpler.

That said, I think we’re still in the “GPT-3.5” phase of image editing: amazing compared to what came before, but still tripping over repeated patterns (keyboards, clocks, Go boards, hands) and sometimes refusing edits due to safety policies. The gap between hype demos and reproducible results is also very real; I’ve seen outputs veer from flawless to poor with just a tiny prompt tweak.


This item has no comments currently.