Inconsistent Rendering of Japanese Green Peppers in Mixed Vegetable Scenes

Dear Adobe Firefly Team,

I’d like to report a consistent issue I’ve encountered when generating images involving both green bell peppers (specifically, the small Japanese-style "piman") and red chili peppers in the same frame.

While Firefly can successfully generate a realistic Japanese green bell pepper when prompted alone, it fails to maintain the correct shape and identity of the green pepper once red chili peppers are introduced into the same image. In such cases, the green peppers are often rendered as long, thin, or chili-like — which is not representative of the small, plump, lobed green peppers commonly found in Japan.

I've tried using very specific prompts — including clear physical descriptions, ingredient exclusions, and references to “piman” or “Japanese supermarket green peppers” — but the model consistently collapses the distinction between the two types of peppers when both are present.

This significantly limits creative use cases where accurate visual contrast between ingredients is essential.

I understand that image models must generalize based on training data, but I hope future improvements can better handle coexisting, visually distinct objects, especially in contexts where regional variation matters (e.g. Japanese piman vs Western bell pepper or chili pepper).

Thank you for your continued work. Firefly has great potential, and I look forward to seeing it evolve.

Best regards, YT.