I've been testing the image output locally for the past couple of days, and the final result seems to be getting better and better. Previously, I felt the loading time was too long, so I asked @antigravity to optimize the generation time. One version downgraded the model to Gemini 2.5 Flash Image, which was indeed faster, and while the effect wasn't significantly affected, it still felt like something was missing. After checking the current model again, I switched back to Gemini 3 Pro Image, and the effect was definitely much better! It's certainly expensive and slow 🤣, but it looks great! After completing the porting to AI Studio, I plan to specifically test the display effect of different types of visual models. I want to see how to find a cost-effective way to optimize only the prompt, since the credits from Google are about to run out. 🥲🥲🥲 Video effect from @grok (I also recently tested the video generation effect, and let me just say it's...unsatisfactory 🤣)