Feature Request: Reasoning / Interleaved Generation like Nano Banana

Flux is by far the best open-weight Image Generation model. Recently Google Nano Banana / ImageGen 4 (and then ByteDance Seedream 4.0) is the top of Image Generation leaderboard. Nano Banana uses new paradigm "**Interleaved Generation**", something similar to LLM **reasoning** / test-time-compute, make it image generation "**thinking / CoT**" to break down complex prompts to step-by-step edits, resulting in impressive character consistency via multiple prompts / edits. It's like **multi-turn, context aware, CoT**, image generation. (refer to DeepMind interview https://www.youtube.com/watch?v=H6ZXujE1qBA) Nano Banana also leverages prior world knowledge from Gemini to generate and edit images. It's unclear if it's a LLM + Image Gen model, or just another MM-DiT model. I'm just curious, does Black Forest Labs have plan to support similar features in next Flux model architecture? Really appreciate!